Beruflich Dokumente
Kultur Dokumente
Editorial Preface
From the Desk of Managing Editor…
It may be difficult to imagine that almost half a century ago we used computers far less sophisticated than current
home desktop computers to put a man on the moon. In that 50 year span, the field of computer science has
exploded.
Computer science has opened new avenues for thought and experimentation. What began as a way to simplify the
calculation process has given birth to technology once only imagined by the human mind. The ability to communicate
and share ideas even though collaborators are half a world away and exploration of not just the stars above but the
internal workings of the human genome are some of the ways that this field has moved at an exponential pace.
At the International Journal of Advanced Computer Science and Applications it is our mission to provide an outlet for
quality research. We want to promote universal access and opportunities for the international scientific community to
share and disseminate scientific and technical information.
We believe in spreading knowledge of computer science and its applications to all classes of audiences. That is why we
deliver up-to-date, authoritative coverage and offer open access of all our articles. Our archives have served as a
place to provoke philosophical, theoretical, and empirical ideas from some of the finest minds in the field.
We utilize the talents and experience of editor and reviewers working at Universities and Institutions from around the
world. We would like to express our gratitude to all authors, whose research results have been published in our journal,
as well as our referees for their in-depth evaluations. Our high standards are maintained through a double blind review
process.
We hope that this edition of IJACSA inspires and entices you to submit your own contributions in upcoming issues. Thank
you for sharing wisdom.
Managing Editor
IJACSA
Volume 10 Issue 9 September 2019
ISSN 2156-5570 (Online)
ISSN 2158-107X (Print)
©2013 The Science and Information (SAI) Organization
(i)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Editorial Board
Editor-in-Chief
Associate Editors
Chao-Tung Yang
Department of Computer Science, Tunghai University, Taiwan
Domain of Research: Software Engineering and Quality, High Performance Computing, Parallel and Distributed
Computing, Parallel Computing
Elena SCUTELNICU
“Dunarea de Jos" University of Galati, Romania
Domain of Research: e-Learning, e-Learning Tools, Simulation
Krassen Stefanov
Professor at Sofia University St. Kliment Ohridski, Bulgaria
Domains of Research: e-Learning, Agents and Multi-agent Systems, Artificial Intelligence, Big Data, Cloud
Computing, Data Retrieval and Data Mining, Distributed Systems, e-Learning Organisational Issues, e-Learning
Tools, Educational Systems Design, Human Computer Interaction, Internet Security, Knowledge Engineering and
Mining, Knowledge Representation, Ontology Engineering, Social Computing, Web-based Learning Communities,
Wireless/ Mobile Applications
Maria-Angeles Grado-Caffaro
Scientific Consultant, Italy
Domain of Research: Electronics, Sensing and Sensor Networks
T. V. Prasad
Lingaya's University, India
Domain of Research: Intelligent Systems, Bioinformatics, Image Processing, Knowledge Representation, Natural
Language Processing, Robotics
(ii)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
CONTENTS
Paper 1: Artificial Intelligence Chatbots are New Recruiters
Authors: Nishad Nawaz, Anjali Mary Gomes
PAGE 1 – 5
Paper 4: Smartphone Image based Agricultural Product Quality and Harvest Amount Prediction Method
Authors: Kohei Arai, Osamu Shigetomi, Yuko Miura, Satoshi Yatsuda
PAGE 24 – 29
Paper 5: Implementing a Safe Travelling Technique to Avoid the Collision of Animals and Vehicles in Saudi Arabia
Authors: Amr Mohsen Jadi
PAGE 30 – 38
Paper 6: A Compact Broadband Antenna for Civil and Military Wireless Communication Applications
Authors: Zaheer Ahmed Dayo, Qunsheng Cao, Yi Wang, Saeed Ur Rahman, Permanand Soothar
PAGE 39 – 44
Paper 9: A Readiness Evaluation of Applying e-Government in the Society: Shall Citizens begin to Use it?
Authors: Laith T. Khrais, Yara M. Abdelwahed, Mohammad Awni Mahmoud
PAGE 55 – 59
Paper 10: Generating and Analyzing Chatbot Responses using Natural Language Processing
Authors: Moneerh Aleedy, Hadil Shaiba, Marija Bezbradica
PAGE 60 – 68
Paper 11: The Criteria for Software Quality in Information System: Rasch Analysis
Authors: Wan Yusran Naim Wan Zainal Abidin, Zulkefli Mansor
PAGE 69 – 75
Paper 12: Chemical Reaction Optimization Algorithm to Find Maximum Independent Set in a Graph
Authors: Mohammad A. Asmaran, Ahmad A. Sharieh, Basel A. Mahafzah
PAGE 76 – 91
(iii)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Paper 13: DLBS: Decentralize Load-Balance Scheduling Algorithm for Real-Time IoT Services in Mist Computing
Authors: Hosam E. Refaat, Mohamed A.Mead
PAGE 92 – 100
Paper 14: Decision Making Systems for Managing Business Processes in Enterprises Groups
Authors: Ali F. Dalain
PAGE 101 – 108
Paper 15: An Extended Consistent Fuzzy Preference Relation to Evaluating Website Usability
Authors: Tenia Wahyuningrum, Azhari Azhari, Suprapto
PAGE 109 – 116
Paper 16: Socialization of Information Technology Utilization and Knowledge of Information System Effectiveness at
Hospital Nurses in Medan, North Sumatra
Authors: Roymond H. Simamora
PAGE 117 – 121
Paper 18: Customers Churn Prediction using Artificial Neural Networks (ANN) in Telecom Industry
Authors: Yasser Khan, Shahryar Shafiq, Abid Naeem, Sheeraz Ahmed, Nadeem Safwan, Sabir Hussain
PAGE 132 – 142
Paper 19: Modified Seam Carving by Changing Resizing Depending on the Object Size in Time and Space Domains
Authors: Kohei Arai
PAGE 143 – 150
Paper 20: A Novel Approach to Rank Text-based Essays using Pagerank Method Towards Student’s Motivational Element
Authors: M Zainal Arifin, Naim Che Pee, Nanna Suryana Herman
PAGE 151 – 158
Paper 22: The Model of Game-based Learning in Fire Safety for Preschool Children
Authors: Nur Atiqah Zaini, Siti Fadzilah Mat Noor, Tengku Siti Meriam Tengku Wook
PAGE 167 – 175
Paper 24: Investigation of Pitch and Duration Range in Speech of Sindhi Adults for Prosody Generation Module
Authors: Shahid Ali Mahar, Mumtaz Hussain Mahar, Shahid Hussain Danwar, Javed Ahmed Mahar
PAGE 187 – 195
Paper 25: Classification of C2C e-Commerce Product Images using Deep Learning Algorithm
Authors: Herdian, Gede Putra Kusuma, Suharjito
PAGE 196 – 203
(iv)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Paper 26: Design and Learning Effectiveness Evaluation of Gamification in e-Learning Systems
Authors: Mohammad T. Alshammari
PAGE 204 – 208
Paper 28: A Nested Genetic Algorithm for Mobile Ad-Hoc Network Optimization with Fuzzy Fitness
Authors: NourElDin S. Eissa, Ahmed Zakaria Talha, Ahmed F. Amin, Amr Badr
PAGE 222 – 228
Paper 30: Strategic Planning towards Automation of Fiber To The Home (FTTH) Considering Optic Access Network (OAN)
Model
Authors: Abid Naeem, Shahryar Shafique, Zahid Wadud, Sheeraz Ahmed, Nadeem Safwan, Zeeshan Najam
PAGE 234 – 244
Paper 31: Performance Evaluation of Different Data Mining Techniques for Social Media News Credibility Assessment
Authors: Sahar F. Sabbeh
PAGE 245 – 256
Paper 33: Microcontroller-based Vessel Passenger Tracker using GSM System: An Aid for Search and Rescue Operations
Authors: Joel I. Miano, Ernesto E. Empig, Alexander R. Gaw, Ofelia S. Mendoza, Danilo C. Adlaon, Sheena B.
Cañedo, Roan Duval A. Dangcal, Angelie S. Sumalpong
PAGE 261 – 268
Paper 34: Utilizing Feature Selection in Identifying Predicting Factors of Student Retention
Authors: January D. Febro
PAGE 269 – 274
Paper 35: An Enhanced Deep Learning Approach in Forecasting Banana Harvest Yields
Authors: Mariannie A Rebortera, Arnel C Fajardo
PAGE 275 – 280
Paper 36: Developing a Dengue Forecasting Model: A Case Study in Iligan City
Authors: Ian Lindley G. Olmoguez, Mia Amor C. Catindig, Minchie Fel Lou Amongos, Fatima G. Lazan
PAGE 281 – 286
Paper 37: Performance Evaluation of Network Gateway Design for NoC based System on FPGA Platform
Authors: Guruprasad S.P, Chandrasekar B.S
PAGE 287 – 292
(v)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Paper 38: Wireless Multimedia Sensor Networks based Quality of Service Sentient Routing Protocols: A Survey
Authors: Ronald Chiwariro, Thangadurai. N
PAGE 293 – 299
Paper 39: Embedded System Interfacing with GNSS user Receiver for Transport Applications
Authors: Mohmad Umair Bagali, Thangadurai. N
PAGE 300 – 308
Paper 40: Empirical Performance Analysis of Decision Tree and Support Vector Machine based Classifiers on Biological
Databases
Authors: Muhammad Amjad, Zulfiqar Ali, Abid Rafiq, Nadeem Akhtar, Israr-Ur-Rehman, Ali Abbas
PAGE 309 – 318
Paper 41: Computer-based Approach to Detect Wrinkles and Suggest Facial Fillers
Authors: Amal Alrabiah, Mai Alduailij, Martin Crane
PAGE 319 – 325
Paper 44: Enhancing Visualization of Multidimensional Data by Ordering Parallel Coordinates Axes
Authors: Ayman Nabil, Karim M. Mohamed, Yasser M. Kamal
PAGE 340 – 344
Paper 46: A Comparison Review based on Classifiers and Regression Models for the Investigation of Flash Floods
Authors: Talha Ahmed Khan, Muhammad Alam, Kushsairy Kadir, Zeeshan Shahid, M.S Mazliham
PAGE 352 – 359
Paper 48: Internal Threat Defense using Network Access Control and Intrusion Prevention System
Authors: Andhika Surya Putra, Nico Surantha
PAGE 371 – 375
Paper 49: CBRm: Case based Reasoning Approach for Imputation of Medium Gaps
Authors: Anibal Flores, Hugo Tito, Carlos Silva
PAGE 376 – 382
Paper 50: Performance Impact of Relay Selection in WiMAX IEEE 802.16j Multi-hop Relay Networks
Authors: Noman Mazhar, Muhammad Zeeshan, Anjum Naveed
PAGE 383 – 391
(vi)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Paper 51: Evaluating Factors for Predicting the Life Dissatisfaction of South Korean Elderly using Soft Margin Support
Vector Machine based on Communication Frequency, Social Network Health Behavior and Depression
Authors: Haewon Byeon, Seong-Tae Kim
PAGE 392 – 398
Paper 52: How to Improve the IoT Security Implementing IDS/IPS Tool using Raspberry Pi 3B+
Authors: Ruíz-Lagunas Juan Jesús, Antolino-Hernández Anastacio, Reyes-Gutiérrez Mauricio René, Ferreira-
Medina Heberto, Torres-Millarez Cristhian, Paniagua-Villagómez Omar
PAGE 399 – 405
Paper 53: Intrusion Detection System based on the SDN Network, Bloom Filter and Machine Learning
Authors: Traore Issa, Kone Tiemoman
PAGE 406 – 412
Paper 54: Development of a Vehicle for Driving with Convolutional Neural Network
Authors: Arbnor Pajaziti, Xhevahir Bajrami, Fatjon Beqa, Blendi Gashi
PAGE 413 – 420
Paper 57: A New Shoulder Surfing and Mobile Key-Logging Resistant Graphical Password Scheme for Smart-Held
Devices
Authors: Sundas Hanif, Fahad Sohail, Shehrbano, Aneeqa Tariq, Muhammad Imran Babar
PAGE 432 – 437
Paper 58: Deep CNN-based Features for Hand-Drawn Sketch Recognition via Transfer Learning Approach
Authors: Shaukat Hayat, Kun She, Muhammad Mateen, Yao Yu
PAGE 438 – 448
Paper 59: A Distributed Approach based on Transition Graph for Resolving Multimodal Urban Transportation Problem
Authors: Mohamed El Moufid, Younes Nadir, Khalid Boukhdir, Siham Benhadou, Hicham Medromi
PAGE 449 – 454
Paper 60: An Intelligent Semi-Latin Square Construct for Measuring Human Capital Intelligence in Recruitment
Authors: Emmanuel C.Ukekwe, Francis S. Bakpo, Mathew C.Okoronkwo, Gregory E.Anichebe
PAGE 455 – 461
Paper 61: Human Gait Feature Extraction based-on Silhouette and Center of Mass
Authors: Miftahul Jannah, Sarifuddin Madenda, Tubagus Maulana Kusuma, Hustinawaty
PAGE 462 – 468
Paper 62: Computer Simulation Study: An Impact of Roadside Illegal Parking at Signalised Intersection
Authors: Noorazila Asman, Munzilah Md Rohani, Nursitihazlin Ahmad Termida, Noor Yasmin Zainun, Nur Fatin
Lyana Rahimi
PAGE 469 – 475
(vii)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Paper 63: Assessment of IPv4 and IPv6 Networks with Different Modified Tunneling Techniques using OPNET
Authors: Asif Khan Babar, Zulfiqar Ali Zardari, Nazish Nawaz Hussaini, Sirajuddin Qureshi, Song Han
PAGE 476 – 482
Paper 64: Security and Privacy Awareness: A Survey for Smartphone User
Authors: Md. Nawab Yousuf Ali, Md. Lizur Rahman, Ifrat Jahan
PAGE 483 – 488
Paper 65: Support Vector Machine for Classification of Autism Spectrum Disorder based on Abnormal Structure of Corpus
Callosum
Authors: Jebapriya S, Shibin David, Jaspher W Kathrine, Naveen Sundar
PAGE 489 – 493
Paper 66: IoT based Temperature and Humidity Controlling using Arduino and Raspberry Pi
Authors: Lalbihari Barik
PAGE 494 – 502
Paper 67: The Use of Geospatial Technology for Epidemiological Chagas Analysis in Bolivia
Authors: Natalia I. Vargas-Cuentas, Alicia Alva Mantari, Avid Roman-Gonzalez
PAGE 503 – 514
Paper 68: A Novel Secure Fingerprint-based Authentication System for Student’s Examination System
Authors: Abdullah Alshbtat, Nabeel Zanoon, Mohammad Alfraheed
PAGE 515 – 519
Paper 69: Ensemble and Deep-Learning Methods for Two-Class and Multi-Attack Anomaly Intrusion Detection: An
Empirical Study
Authors: Adeyemo Victor Elijah, Azween Abdullah, NZ JhanJhi, Mahadevan Supramaniam,
Balogun Abdullateef O
PAGE 520 – 528
Paper 70: Timed-Arc Petri-Nets based Agent Communication for Real-Time Multi-Agent Systems
Authors: Awais Qasim, Sidra Kanwal, Adnan Khalid, Syed Asad Raza Kazmi, Jawad Hassan
PAGE 529 – 537
Paper 71: Scale and Resolution Invariant Spin Images for 3D Object Recognition
Authors: Jihad H’roura, Aissam Bekkari, Driss Mammass, Ali Bouzit, Patrice M´eniel, Alamin Mansouri, Micha¨el
Roy
PAGE 538 – 545
Paper 72: A Novel Approach for Ontology-Driven Information Retrieving Chatbot for Fashion Brands
Authors: Aisha Nazir, Muhammad Yaseen Khan, Tafseer Ahmed, Syed Imran Jami, Shaukat Wasi
PAGE 546 – 552
(viii)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Paper 74: Prediction of Academic Performance Applying NNs: A Focus on Statistical Feature-Shedding and Lifestyle
Authors: Shithi Maitra, Sakib Eshrak, Md. Ahsanul Bari, Abdullah Al-Sakin, Rubana Hossain Munia, Nasrin Akter,
Zabir Haque
PAGE 561 – 570
Paper 76: Thai Agriculture Products Traceability System using Blockchain and Internet of Things
Authors: Thattapon Surasak, Nungnit Wattanavichean, Chakkrit Preuksakarn, Scott C.-H. Huang
PAGE 578 – 583
Paper 77: Mobile Agent Platform based Wallet for Preventing Double Spending in Offline e-Cash
Authors: Irwan, Armein Z. R. Langi, Emir Husni
PAGE 584 – 591
Paper 78: Towards A Proactive System for Predicting Service Quality Degradations in Next Generation of Networks based
on Time Series
Authors: Errais Mohammed, Rachdi Mohamed, Al Sarem Mohammed, Abdel Hamid Mohamed Emara
PAGE 592 – 598
(ix)
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The purpose of the paper is to assess the artificial functional areas, this will minimize the time and effort of
intelligence chatbots influence on recruitment process. The human resources, in other words artificial intelligence (AI)
authors explore how chatbots offered service delivery to attract will replace human routine work, enforced them to generate
and candidates engagement in the recruitment process. The aim strategies and become craft in the domain.
of the study is to identify chatbots impact across the recruitment
process. The study is completely based on secondary sources like Artificial intelligence (AI) chatbots developed to make
conceptual papers, peer reviewed articles, websites are used to messages to provide assistants to the consumers for 24/7, to
present the current paper. The paper found that artificial answer all queries and acting like FB messenger, webchat, but
intelligence chatbots are very productive tools in recruitment the competitive environment enthusiastically looking for new
process and it will be helpful in preparing recruitment strategy added features in artificial intelligence (AI) chatbots to handle
for the Industry. Additionally, it focuses more on to resolve all the raised complex problems, therefore artificial
complex issues in the process of recruitment. Through the intelligence (AI) chatbots are much demanded in chatbot
amalgamation of artificial intelligence recruitment process is market. Additionally, chatbots present organization to be data
increasing attention among the researchers still there is driven and pivotal in the success of the business.
opportunity to explore in the field. The paper provided future
research avenues in the field of chatbots and recruiters. The paper is structured as follows, the literature is
reviewed, the purpose of the study and then discussion is
Keywords—Artificial intelligence; chatbots; recruitment presented. The paper dismisses with a conclusion and ideas for
process; candidates experiences; employer branding tool; future research studies.
recruitment industry
II. LITERATURE REVIEW
I. INTRODUCTION
According to [5] recruitment process enhance quality with
In the new phenomenon of information technology and AI, it will assist employers to select suitable candidate with in
human resource management decades of decades have a second to ensure whether the candidate is suitable or not. AI
observed that, the embedded information technology and has constantly developed over the time to provide deeper
human resource management in new term as a human resource insights. This will ensure the organizations not only hire, the
information system, digital human resource management, right candidate for the organization but also with right skill.
automation human resource, enterprise resource planning IOT [6] main aim of the study is to explain digital technologies
(Internet of Things), data mining, [1], [2], [3], [4] and freshly influence HR organization. This paper throws light on how
artificial intelligence added to the old wine. Again, we can digital technologies have reconfigured the HR organization as
spot the vigor of information technology innovations in well as digital technologies transformation and support to the
business. The latest trouble shooters (solutions) developed by organizational effectiveness, talent strategy execution,
technology to the complex issues of all the various functions succession planning, monitoring operations, transactional case
of organizations are drawing more attention of the managers monitoring, recruitment process (robotics automation,
from different departments, areas and domains, not excluding artificial intelligence (chatbots), [7], workforce planning,
the human resource department. analytics, employee banding management, rewards and
The term artificial intelligence (AI), commonly used for engagement, employee relations and effectiveness of
software, machines, system and computers. First time, in the organization and finally concluded that digital technology
era of industrial revolution, Rossum’s Universal Robots forcing HR process and organization structure into the
(R.U.R) brought into picture by Czech Karel and it is named transformation and [8] through AI can identify right talent
as ROBOT. But, in the case of artificial intelligence term has leaders, they may easily procure deep insights of training
been introduced by John McCarthy (1956) appeared in the needs, how to cradle time consuming in screening the
academic conference and explained the term, as he suggested resumes, unbiased candidate screening and helpful in
artificial intelligence (AI) will contribute in future in the analyzing personality traits and finally concluded that conduct
following specializations, like applied science, psychology, onboarding via chatbots, this will enhance workforce
medical sciences, linguistics, biology, engineering and experience. AI is significantly improving in HR functions,
interdisciplinary programs. especially in performance management. Traditionally HR
managers are evaluating employees performance once in a
Firstly, we need to know, why HR emerging artificial year, but AI-driven performance management removing
intelligence (AI), because companies want to extend their unwanted delay in measurement of performance via real time
business operations to gain profit and new ventures across the points, face to face communication, chatbots and these
globe, it is possible via new technology. Therefore, the technologies providing daily performance, this will prioritize
organizations aiming to adopt automation process across the retention across HR [9].
1|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Artificial intelligence understands the human language and Internet search engines through Google, Google scholar were
not only command but learning from human conversion, utilized to identify and access the relevant working papers,
transforming like intelligent agent, chatbot is a computer reports, blogs and presentations were used to ensure
program the conversations through an auditory or methods of comprehensive coverage of the literature.
textual via natural language processing (NLP), natural
language understanding (NLU) and natural language IV. RESULTS
generation (NLG) for interacting humans like ALICE. [10]. The results identified that there is an increase in the
An automated mediator stimulate job-seeker to ask technology development in human resource management,
questions directly with recruiter about the salary, incentives, especially in the recruitment process that will have more
leave facilities, FAQ, workforce diversity, complex queries influence in the future. In past years companies used various
and other related questions [11]. Another piece of work done technologies for the recruitment such as social media,
in this direction chatbots eliminate the routine work in the employee portals, job portals, internal and external networks,
recruitment process, the RobRecruiters chatbots are social networking, peer referral, emails, mobile messages, cell
automating end to end recruitment process and doing regular phone communications are used to attract star performers, best
process of attendance tracking, goal tracking, reviews of talents for their teams to perform better in the organisation.
performance, surveys related employees, balance leaves and In the present transparent digital era, the recruiters, are
other activities, enables the HR managers and HR team to more experienced because the candidates are more tech-savvy
move towards success to place organization in top in in the mobile environment, and it reinforces the employment
digitalized era. [12]. brand, an organization employment brand is extremely
(Adams, 2018) chatbots are transforming and assisting in important, it will attract best performers into their talent pool.
recruitment process to know candidate experience, for Therefore, organizations always want to manage their
effective communication between candidate and recruiter, employment brand to pull candidate towards them.
questions and answers, in identification of qualified candidate There are many new implications, solutions and innovative
and to schedule conversation and finally chatbots taking all ideas around cognitive technologies, for instance, artificial
requirements from the candidate before his/her entry into the intelligence, natural language processing, natural language
organization. The modern recruiters having more additional understanding, natural language generation, machine learning,
responsibilities to have strategies to meet the assigned predictive algorithms and robotics process automation, after
business goals, keep tracking of competitors, keeping realistic introduction of Olivia, the chatbots become very popular in
challenges via benchmark, tracking of employees’ satisfaction the recruitment market and it is providing the guidelines to the
level in all the stages, for this chatbots are useful, because candidates, answer their questions. Moreover, facilitating
there is a daily conversation database, this will be useful to sequence questions to the applicant to answer.
resolve complex issues Joshi (2019).
The above technologies adopt smart methods for collecting
The artificial intelligence empowered chatbots to execute data of the candidate to make progress in various techniques
human conversation in messaging, the unique way of using [15] to identify possible candidate to apply, responding to the
words, shorthand, emotions, [13] at the end it will work on the unsuccessful candidate, screening the candidate, in sending
basis of natural language to support conversation process [14], job offer, and bringing selected candidate into company [16]
the studies are rare in the field of recruitment process. The in other way connecting & collecting all the information of
authors claim that our understanding of artificial intelligence employee into single database and it reveals a new insights
in recruitment process would benefit more intensive by across for better candidate profile to hire and improve the
all levels of recruitment process. In order to develop better effectiveness of the recruitment process [17].
understanding of the recruitment process in human resources
organizations in general, at the same time to have Against this reflection, the aim is to study AI chatbots
effectiveness in different areas of recruitment. Therefore, the impact in the recruitment process. More precisely, the present
authors have proposed the study of artificial intelligence study attempts, AI chatbots, significance across the
chatbots influence in recruitment process. recruitment process function. These are discussed below.
2|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
generally affirm which applicants are appropriate amid the persuade the applicant to click on the company’s career page
first round of pre-screening calls; while we do know this and submit their application as well. The solution is very
works, this can be a long procedure which needs revising. simple in the chatbots. By using an automated facebook
Chatbots are an extraordinary arrangement which can make messenger recruitment chatbot, any potential applicant could
this phase of the evaluation unmistakably increasingly be prompted to opt-in for job alerts and provide them with
effective! By conveying a text message to every potential their facebook profile information, as well as showing them
applicant that prompts a progression of short, pre- job openings, information about the process of application or
characterized questions, the appropriate applicants can be even videos about working at the company. The possibility of
effectively be sifted through from the unsatisfactory. submitting the application by the candidate increases many
folds if they are engaged right from the point of initial interest
Recruiters’ can convey several texts in a few minutes and as they’ve already had contact and established a rapport with
get reply speedily, as compared to the days or even weeks that the company
telephone calls and messages may take.
The shocking fact is that as per Jobvite Recruiting Funnel
B. Get the Right Data at the Right Time of the Right 2017 only 8.52 per centage of visitors to a career site literally
Candidate complete their application. Which means more than 91 per
All organizations dream of updating candidates’ database centage of potential candidates just leaves the site without
every day, replying to clients faster as well as creating a long leaving any information for the company, hence even if the
term relationship with the candidates. Its not a dream company wants to contact them in the future they cannot due
anymore, it is now possible with chatbot. Companies can to lack of information. This can be a very big problem to
deploy a chatbot connected to the database to regularly update companies trying to attract talented and skilled people for their
it. Companies can check their database information which is a job vacancies. To tackle this issue a chatbot can help by
key influence. With individual and automated discussions engaging candidates through a messenger, replying to any
companies can refresh applicants’ accessibility, their present questions regarding any misgivings they have which might
position, their mobility or even a new certification. Companies daunt them from submitting any application, it can also give
can add value to their database by deploying a chatbot within reminder to candidates to complete all the information in case
the preferred audience which can add value to the database. it is insufficient.
C. Qualifying Candidates F. Question and Answer (FQA)
Chatbots can perform an excellent job of filtering out the Before applying the applicants needs to know about the
good from the bad by asking questions to the applicants job, the company and various requirements and it is extremely
related to skills, qualifications, and past experiences which can frustrating if a candidate has to look for all these answers on
be otherwise a tedious and time consuming task for the an FAQ page. This can be completely changed with
recruiters. It can then effectively rank and qualify a whole implementation of chatbots. Companies can use chatbots to
group of candidates in terms with the required criteria of the answer FAQ by recognizing keywords mentioned by the
organization. applicant. It's crucial to make the answers understandable and
informative which will enable the applicant to acquire
With all the extensive administrative tasks that come with
knowledge and clear understanding promptly leading to
hiring a candidate, Chatbots can take off a lot of the load by
applicant’s satisfaction.
setting up inevitable calls and scheduling meetings keeping in
mind both the parties. Apparently, these stages of the process It is less intimidating communicating to a chatbot where an
require the human but chatbot ensures all requirements are applicant can be just themselves and stay calm and composed.
addressed before humans take over. Chatbots can respond to the applicants in a user-friendly way
and if the answer is not known it will refer the applicant to the
D. Get more Qualified Appliers into Job Offers
right person to create a contented experience. This is the age
Companies are aggressively forwarding emails to their of data driven decisions hence a chatbots could be linked to a
database to get more job applicants however the achievement platform which can gather important data. This type of
rate isn't as high as anticipated. With a chatbot on messaging platform will enable the company to tag how many times a
apps companies can draw in the applicants database and push particular question has been asked and what candidates want
them towards the right job at the opportune time. Applicants and are looking for. Chatbots are great assistance to recruiters
can apply without any difficulty through the chatbot. with their prompt replies and instant availability.
There is no signing in required to go on a website nor have G. Responding to the Unsuccessful
they to go through tedious application process. The job
According to Eyal Grayevsky, Mya “applicants who never
applicants will receive job offers on their messaging apps
get to hear back after submitting their applications from the
rather than receiving it through emails. At that point, they can
recruiter are approximately 85 per centage” this results in poor
apply for them without changing to another application or site.
applicant experience and a wrong impression of the company.
E. Increased Numbers of Applications This can also lead to loosing potential candidates which might
Due to the impact of social media, facebook recruitment be a critical issue, as they will not be motivated to reapply in
via facebook groups, posts are becoming extremely popular in the same company, who might be better suited for another
attracting new candidates. However, the problem is to position in that company in the future.
3|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
To remain competitive companies must attract the best high volume, it very well may be a major errand for a human
talent and skilled professionals. The companies can be to deal with. Proficiently that is, chatbot can intercede with
successful in this only if they are seen as someone who treats faster reply and speed up the procedure.
people with respect. With numerous numbers of applications
for one job position it is practically not possible for recruiters To emerge of the commotion, recruiters ought to possibly
to personally reject everyone, hence it makes the chatbots the change their recruitment strategy and make it more applicant
most suitable option. Chatbots can promptly respond to the driven. The time that applicants take to send resumes and
applicants once a decision is taken. The applicants recruiters hit them up ought to be short. Chatbots can get this
acknowledge this kind of swiftness and at the same time they going. Chatbots can return to the applicants quickly, making
don’t develop any negative feeling towards the organization as the applicants as well as the recruiters contented on that front.
they don’t go through the anxiety of waiting time. With a correct approach companies can make the job search
experience where conversation is concerned a more natural
H. Screening Candidates Application process.
Chatbots can initiate a conversation with the applicants In the present day’s marketing recruitment the experience
once they apply on a company’s job site. Chatbots may ask of the applicants is becoming extremely important. The
many questions while communicating with them. The applicants’ journey must be effortless, straightforward, and
questions can range from asking about work experience, inviting and this must be ensured by the talent acquisition
previously where they have worked, their areas of interest and leaders. Recruitment chatbots can connect with applicants in a
so on. When this process is over the chatbots assess the conversational trade as well as answer recruiting FAQs, a
applicant for relevancy of the unfilled position. The decision is boundary that prevents numerous applicants from applying.
taken by the recruitment chatbot based on the conversation With recruiting web chat arrangement like career chat,
that took place, resume details and assessing the job applicants can study the organization and draw in recruiters in
requirement if the candidate is best bit for the job. live agent modes and computerized modes.
I. Assess Candidates’ Recruitment Experience M. Candidate Experience Feedback
Companies must be aware about how the applicants feel Huge number of applicants have poor encounters while
regarding their recruiting experience. Its important to presenting their applications and resumes on the web. As a
differentiate from the competitors which is possible by getting rule, they don't get advised about whether an organization they
feedback and this way a strong relationship can be built. To have connected for has gotten their documents. This
enhance the retention rate of the company its important to ask vulnerability combined with the distress of sitting tight for
the applicants about how they feel. The chatbots can assist in input makes a negative impact on an organization's validity.
checking candidate’s feedback after interviews and get deep
insights about how companies create a bond with people Through the incorporation of selecting artificial
within the talents pool and useful in recruitment strategy as intelligence, many applicants can be obliged promptly and
well. advised with the outcomes of their interview once it's finished.
This brings down their nervousness and encourages them
J. Interview Scheduling of the Candidates proceed onward to discovering openings that are more
Scheduling the interview with the candidate is another appropriate for them. Organizations that make HR bots are
time-consuming task. Intellectual chatbots are capable of looking for better approaches to improve their process to
accessing the calendar of the recruiters to check if they are fulfill the requirements of clients around the world.
available and then schedule the date and time for the relevant Job pal perceives the requirement for organizations to
candidate. In today’s time it’s not very much effective to just begin connecting with applicants the minute they apply for
make calls to the candidate as most of them don’t answer to work/ job. With that, they have fabricated AI-controlled
unrecognized phone numbers. Also, it might be bothersome to chatbots to mechanize the correspondence among employees
call the candidate when they are working with their current and applicants accordingly accelerating the procuring
company or they have to request for a convenient time for procedure.
both the parties. This whole process could be very time
consuming. But for a chatbot this task will not be tedious, and N. Notation Feedback
they are great in this type of repetitive tasks. Structured inquiries that accompany predefined answers
K. Enhance Candidates On-Boarding will aid in real-time feedback. Instead of using forms the
applicants can have the privilege of dynamic interactive
The very first step for a candidate is on-boarding in an conversational interface to share their feedback. In the point of
organization which is also a long and key process. To improvement, Questions from the employers can be put
smoothen the process the organizations can deploy a chatbot forward to candidates such as, “where do you think we need to
to deliver significant information at the right time to the newly improve” to get an understanding of areas of improvement.
recruited employees. The feedback of the applicants could be shared with the
L. Candidate Experience recruiters to initiate necessary actions and filter the process to
It's nothing astonishing that one will need to get a chatbot best fit the human capital needs. Using these observations the
engaged with this piece of the procedure. With a new job companies can make sure of providing an engaging experience
opening numerous numbers of applicants will be pulled in a to candidates and help HRs in eradicating human bias and the
possibility for any error.
4|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
5|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The development of the microprocessor industry in by the strict separation of the processing and memory resources
terms of speed, area, and multi-processing has resulted with in the computer system. In such processor-centric system the
increased data traffic between the processor and the memory in a memory is used for storing data and programs, while the
classical processor-centric Von Neumann computing system. In processor interprets and executes the program instructions in a
order to alleviate the processor-memory bottleneck, in this paper sequential manner, repeatedly moving data from the main
we are proposing a RISC-based memory-centric processor memory in the processor registers and vice versa, [1].
architecture that provides a stronger merge between the Assuming that there is no final solution for overcoming the
processor and the memory, by adjusting the standard memory processor-memory bottleneck, modern computer systems
hierarchy model. Indeed, we are developing a RISC-based
implement different types of techniques for "mitigating" the
processor that integrates the memory into the same chip die, and
thus provides direct access to the on-chip memory, without the
occurrence of this problem, [10], (ex. branch prediction
use of general-purpose registers (GPRs) and cache memory. The algorithms, speculative and re-order instructions execution,
proposed RISC-based memory-centric processor is described in data and instruction pre-fetching, and multithreading, etc.). In
VHDL and then implemented in Virtex7 VC709 Field fact, the most applied method for approaching data closer to
Programmable Gate Array (FPGA) board, by means of Xilinx the processor is the use of multi-level cache memory, as faster,
VIVADO Design Suite. The simulation timing diagrams and but smaller and more expensive data storage than the main
FPGA synthesis (implementation) reports are discussed and memory. Regarding that, the research stated in [11] discusses
analyzed in this paper. that the capacity and the area of on-chip cache memory have
shown steady growth, as a result of the increased number of
Keywords—FPGA; memory-centric computing; processor in on-chip processing cores, which have imposed even greater
memory; RISC architecture; VHDL requirements to the memory system. For example, up to 40%
of the chip area in Intel's 65nm processors is occupied by
I. INTRODUCTION caches, [12], used solely for hiding the memory latency.
The growing technological progress over the last several
Despite the grand popularity of cache memory used in the
decades has caused dramatic improvements in processor
modern computer systems, we should note that each cache
performances, providing speed-up of processor's working
level presents a redundant copy of the main memory data that
frequency, increased number of instructions that can be issued
would not be necessary if the main memory had kept up with
and processed in parallel, [1], [2], multithreading, pre-fetching,
the processor speed. According to [13], cache memory causes
etc. According to Moore's law, [3], [4], the integrated circuits
up to 40% increase of the system's energy consumption,
production technology has enabled doubling of the number of
because it adds extra hardware resources and requires the
transistors on a chip every 18 months, which resulted with the
implementation of complex mechanisms, [14], for maintaining
creation of multi-core processors over the last decade. This
memory consistency. Besides that, the misses in cache memory
trend of processor technology growth has brought performance
bring unpredictability in the timing of the program, which is
improvements on the computer systems, but not for all the
not very suitable for real-time systems.
types of applications, [5]. The reason for such divergence is
due to the bottleneck problem in the communication between On the other hand, the development of some powerful
the processor and the main memory (which is by default placed processor architectures, such as vector, [15], wide superscalar,
out of the processor), caused by the growing disparity of [16], VLIW (very long instruction word), [17], and EPIC
memory and processor speeds, [6]. Therefore, we can say that (explicitly parallel instruction computing), [18], did not
not long ago, off-chip memory was able to supply the achieve the expected success, because of their inability to
processor with data at an adequate rate. Today, with processor provide fast and high throughput access to the memory system.
performances increasing at a rate of about 70 percent per year Considering the difference between the processor and the
and memory latency improving by just 7 percent per year, it memory speeds, we believe that the relatively small number of
takes a dozens of cycles for data to travel between the fast GPRs in the processor is the major obstacle for achieving
processor and the main memory, [7], [8], which is basically high data throughput. This is mainly expected in the case of
placed outside of the processor chip. executing a program that works with larger data set that needs
to be placed into the processor for a short time, but there are
The computer systems that are used today are mainly based
not enough free registers. Examples for such applications are:
on the Von Neumann architecture, [9], which is characterized
6|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
processing of data flows, calculating vast logical-arithmetical forwarding, access to on-chip memory, etc. Section IV presents
expressions or traversing complex data structures, etc. In such simulations and synthesis results from the FPGA implemen-
cases, the high speed of access to GPRs doesn't bring many tation of the proposed RISC-based memory-centric processor.
advantages, because the complete set of required data cannot The paper ends with a conclusion, stated in section V.
be placed into the register set at the proper time. Therefore, the
author of [19] purposes a register-less processor which uses II. CURRENT STATE
only cache memory (inside and outside of the processor) to The extraordinary increase of microprocessor speed has
communicate with the main memory. Additionally, the authors caused significant demands to the memory system, requiring an
of [20] and [21] suggest the use of Scratchpad memory as a immediate response to the CPU (central processing unit)
small software-managed on-chip memory that is separated of requests. Considering that the memory price, capacity, and
the cache memory and can be accessed in a single proc. cycle. speed are in direct opposition, an ideal memory system cannot
A few decades ago, in the 1990ties, some researches be implemented in practice, [2]. Therefore, today's modern
predicted that the memory behavior would be preponderant computer systems are characterized with hierarchical memory,
over the global performances of the computer system. Their organized in several levels, each of them having smaller, faster
proposals suggested the design of "smart memories" that will and more expensive memory, compared to the previous level.
include processing capabilities. Therefore, several memory- The hierarchical approach of memory organization is based
centric approaches of integrating or approaching the memory on the principle of temporal and spatial locality, [1], [2], and
closer to the processing elements have been introduced, the rule "smaller is faster" which states that smaller pieces of
including: computational RAM, [22], Mitsubishi М32R/D, memory are usually faster and hence more expensive than the
[23], DIVA, [24], Terasys, [25], intelligent RAM, [26] - [28], larger ones. According to that, cache memories have lower
parallel processing RAM, [29], DataScalar, [30], and an access time, but on the other hand they bring indeterminism in
intelligent memory system, known as active pages model, [31]. the timing of the program, as a result of the misses that can
Within these memory-centric systems, the processor can be occur during the memory accesses (read or write). This is also
realized as some simple RISC or complex superscalar confirmed with equations 1 and 2, which give the expressions
processor and may contain a vector unit, as is the case with the for computing average memory access time and program
Intelligent RAM. execution time, accordingly. The relation between these
The aim of this paper is to develop a novel RISC-based equations is expressed with the CPI (cycles per instruction)
memory-centric processor architecture, which suggests an parameter, which value depends on the average memory access
integration of processor and memory on the same chip die and time. Therefore, if many misses to intermediate memory levels
proposes removal of general-purpose registers and cache occur, the program's execution time will increase, resulting in
memory (inside and outside of the processor) from the standard many wasted processor cycles.
memory hierarchy. Contrary to the other memory/logic merged Average memory access time =
chips, which mostly use the standard memory hierarchy model
for data access, the proposed RISC-based memory-centric = Hit time + Miss rate * Miss penalty (1)
processor provides direct access to the data into its on-chip Execution time = Instructions number *CPI*Clock period (2)
memory (without the use of explicit LOAD and STORE
instructions) and includes specialized control unit that performs According to the previous assumptions, we can say that
4-stage pipelining of instructions, allowing every (arithmetical, multi-level cache memories can cause reduction of the memory
logical, branch and control) instruction to be completed in а access time, but at the cost of additional hardware complexity,
single tact cycle. If this logic is manufactured as an ASIC increased power consumption, unpredictable program's timing
(application-specific integrated circuit) it cannot be reused for and extra redundancy in the system. Other techniques for
further extensions, so in this paper we are investigating the memory latency reduction include a combination of large
possibilities to utilize a reconfigurable hardware platform - cache memories with some form of branch predictive
Virtex7 VC709 FPGA board, [32]. In that process, we are speculation, or out-of-order execution, [14]. These methods
developing a VHDL model of the proposed RISC-based also increase the chip area and cause extra complexity on both
memory-centric processor, and then we are simulating the the hardware and software level. Even other more complex
functionalities of the proposed processor and analyzing the approaches of computing like vector, wide superscalar, VLIW
characteristics and the complexity of its FPGA and EPIC suffer from low utilization of resources, implemen-
implementation, by means of Xilinx VIVADO Design Suite. In tation complexity, and immature compiler technology, [15] -
fact, FPGA technology is very suitable for the purposes of this [18]. When it comes to processor architectures, we can say that
research since it represents a good compromise between the integration of multiple cores or processors on a single chip
performance, price, and re-programmability, [33]. die brings even greater demands to the memory system,
increasing the number of slow off-chip memory accesses, [8].
The rest of this paper is organized as follows: Section II
gives an overview of different techniques and methods used to In order to tolerate the memory latency and allow the
alleviate the processor-memory bottleneck and also discusses processor to execute other tasks while a memory request is
several memory-centric approaches of computing. Section III being served, a separate group of memory latency tolerance
presents the proposed RISC-based memory-centric processor, techniques was introduced. Some of the most popular methods
describing its basic architectural characteristics, including in this group are multithreading, [2], instruction and data pre-
instruction set, addressing modes, pipelining support, data fetching, [1] and non-blocking caches, [34]. In general, the
7|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
usage of these methods contributes to the "reduction" of the called the Machine, [39], which utilizes non-volatile memory
memory latency, but on the other hand it increases the memory as a true DRAM replacement. A more detailed study about
traffic, leading to a higher instruction and data rate. As a result other proposals for overcoming the processor-memory
of the limited bandwidth on the memory interface, additional bottleneck is presented in our previous research, given in [40].
latency can be generated.
Considering the adjustments of the standard memory
Besides the previously discussed memory latency reduction hierarchy model, presented in some of the previously discussed
and tolerance methods, there are several proposals, which approaches (ex. PERL, Scratchpad, Machine), we can say that
present some modifications into the classic multi-level memory the extension or revision of their work can be a good starting
hierarchy and provide nonstandard faster access to the main point for further research. In that process, we can first perceive
memory. For example, the author of [19] proposes a register- that the relatively small number of fast GPRs in the highest
less processor that performs all the operations directly with the level of the memory hierarchy is the major obstacle for
cache memory, organized in several layers (on-chip and off- achieving high data throughput. After that, we can consider
chip), excluding the explicit use of GPRs. Additionally, the that the cache memory is a limiting factor in real-time
authors of [21] suggest the use of Scratchpad memory as a computing, and is also a redundant memory resource, which
small high-speed on-chip memory that maps into the adds extra hardware complexity and power consumption into
processors address space at a predefined memory address the system. Therefore, our research will continue into the
range. Opposite to the cache memory, the Scratchpad memory direction of developing a novel RISC-based memory-centric
is allocated under software control and is characterized with processor similar to PERL, which will provide direct access to
deterministic behavior, allowing single-cycle access time. This the memory that is integrated into the processor chip, without
small on-chip memory is mostly used for storing in-between the use of GPRs and cache memory. The proposed replacement
results and frequently accessed data, so it requires developing of the two highest memory hierarchy levels with an on-chip
of complex compiler methods for effective data allocation. memory is intended to provide: exclusion of unnecessary data
copying and individual or block data transfer into the GPRs
Contrary to the standard model of processor-centric
and cache memory, a decrease of the capacity of redundant
computing (Von Neumann model), [9], some researchers have memory resources, simplification of the accesses to memory
proposed alternative approaches of memory-centric computing, and removal of complex memory management mechanisms.
which suggests integrating or placing the memory near to the
processor. These proposals are known as computational RAM, III. DESIGN OF RISC-BASED MEMORY-CENTRIC
intelligent RAM, processing in memory chips, intelligent PROCESSOR ARCHITECTURE
memory systems, [22] - [31], etc. These merged memory/logic
chips implement on-chip memory which allows high internal As a referencing point for designing the proposed RISC-
bandwidth, low latency, and high power efficiency, eliminating based memory-centric processor, we make use of a RISC
the need for expensive, high-speed inter-chip interconnects, architecture implementation (MIPS), which is widely applied
[35]. This makes them suitable to perform computations which in the embedded industry and additionally is well documented
require high data throughput and stride memory accesses, such and presented in the leading world's literature in the field of
as FFT, multimedia processing, network processing, etc., [28]. processor architectures. The selected MIPS implementation of
a single-cycle pipelined RISC architecture, presented by D. A.
The integrated on-chip memory in the merged Patterson and J. L. Hennessy in [1], is also used as a basis in
memory/logic chips is usually implemented as SRAM or the PERL processor architecture design. In general, MIPS
embedded DRAM, which is mostly accessed through the processor is characterized with: fix-length instructions, simple
processor's cache memory. Although the processing in/near addressing modes, memory accesses with explicit load and
memory brings latency and bandwidth improvement, still the store instructions, hardwired control unit, large GPR set and
system has to perform unnecessary copying and movement of pipeline operation in five stages (fetch, decode, execute,
data between the on-chip memory, caches, and GPRs. Besides memory access and write back), as shown in Fig. 1.
that, the processing speed, the on-chip memory size, and the
chip cost are limited due to the used implementation According to Fig. 1, a MIPS processor includes: Program
technology and the production process. Moreover, it is even a counter - PC, Instruction Register - IR, pipeline registers, 32
greater challenge to develop suitable compiler support for the general-purpose registers, separated instruction and data cache
system, which will recognize the program parallelism and will memory, 32-bit arithmetical and logical unit, control unit
enable effective utilization of the internal memory bandwidth. (marked with blue), and other selecting and control logic
(multiplexers, decoders, adders, extenders etc.). Therefore,
Having in mind that modern processors are lately dealing MIPS operates only on operands found in its local GPRs,
with both technical and physical limitations, while the memory requiring frequent data transfer between the memory and the
capacity is constantly increasing, it seems that now is the right processor's registers, via load and store instructions. In order to
time to reinvestigate the idea of placing the processor in or near provide easier access and manipulation of memory data, this
to the memory in order to overcome their speed difference, paper proposes a modification and extension of the original
[36] - [38]. A promising approach that targets this problem is pipelined RISC architecture and creation of a novel MEMRISC
presented by the Hewlett Packard international information (Memory Access Reduced Instruction Set Computing)
technology company that suggests novel computer architecture, pipelined processor architecture, shown in Fig. 2.
8|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Instruction[63..56] = Opcode
inoutbus inoutbus inoutbus
WB
Control Unit
EX
DMemWrite
WB
UncondBranchType
0 3
MUX
Shift (CondBranch
Halt Instruction[36..8] 0 2 AND
Left 3
MUX
1 branchFlag) Result
CondBranch, OR Selector[2..0]
1 UncondBranch UncondBranch
ADD
Sign Extend 6
Instruction[55..40] inData
to32bit & BranchFlag
CondBranch ALUOpcode
0
MUX
DMemPTWrite ALUResult
Float
Data ResData
and
DMem DMem
8 Instruction[39..24] Write Read Hazard
MUX
Op1
MUX
in blocks unit
ID/EX
2 Selector
IF/ID
PRsMemWrite[2..0], 6 0
MUX
ShiftType[2],
BRsMemWrite[2..0] ResData ShiftEnable
IMem IMem Op2 1
IMemPTWrite Write Read Imm Type
Integer 3
Instruction Data Instruction[23..8] Type
Extend ShiftOper
Memory Address ShiftOp2
to32bit Selector
Address 0 Op2 Selector
MUX
Instruction 0
MUX
Generator Float
memory 1 1
Extend
segmented Shifter
to32bit 0
MUX
in blocks Instruction[7..3] 0
MUX
Shifter Imm21Shift 1
Instruction[23..3] Constants
Extender ShiftAmmount 1
Load
to32bit 2 Op[2]
Instruction[39..32] LoadResult
Load Imm8 0 PageRegister
Instruction[39..24]
MUX
Constants Imm16 1
BaseRegister
Extender Imm16Up 2
Instruction[39..8] 3
to32bit Imm32 PageReg&BaseReg
Forward
PageReg ExRes
BaseReg
PageReg&BaseReg Forward
AddrResult AddrResult WBRes
Forward_EX
Forward_Wb
ResData
ResAddr
9|P age
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
As shown in Fig. 2, the proposed processor with and thus simplifies the access to the operands, the execution of
MEMRISC architecture uses separated on-chip data and the instructions (pipelining without MEM phase) and the
program memory, instead of GPRs and on-chip cache memory. instruction set (removes explicit LOAD/STORE instructions).
This means that the given processor executes all the operations This way of operation of the MIMOPS processor is managed
on values found in its on-chip memory, avoiding the by a specialized control unit (marked with blue on Fig. 2),
unnecessary and redundant data copying and movements, which provides support for several addressing modes (ex.
which are performed in the MIPS processor, during (load/store) direct, immediate, base, PC-direct, and PC-relative addressing).
data transfers. Therefore if the RISC-based MIPS processor is Generally, the memory operands are addressed directly, while
able to execute a million instructions per second, then the the translation of the virtual addresses to physical addresses is
proposed processor with MEMRISC architecture would be performed via specialized hardware support for virtual memory
able to execute a million instructions on memory operands per that is implemented inside the MIMOPS processor. This refers
second, which is the reason why it is called MIMOPS to segmentation of the on-chip memory and filling it with
processor in continuation. virtual pages, and implementation of page translation tables
and page replacement mechanisms (ex. FIFO).
The proposed MIMOPS processor excludes the GPRs and
the cache memory from the memory hierarchy and thus allows The proposed MIMOPS processor implements separated
direct and simultaneous access to two sources and one result on-chip instruction and data memories that are segmented into
operand, specified in the instruction. These operands are М equal-sized physical blocks (for virtual pages). Each of
selected by a specialized memory address generator unit that is these local memories is organized as an array of N contiguous
used to perform the translation of the input virtual addresses byte-sized elements, whereas each element has a unique
into physical addresses of the paged on-chip memory. Once the physical address. To provide support for address translation
operands are read from the on-chip data memory, the operation and simultaneous access to the separated instruction and data
is executed and then the result is written back to the on-chip on-chip memories, the proposed MIMOPS processor
data memory. In fact, the MIMOPS processor operates in a 4- implements two dedicated hardware units, called instruction
stage pipeline (instruction fetch, instruction decode, execute and data memory address generators. These units translate
and write back), excluding the MEM phase, and allowing every virtual addresses on the fly, performing a look-up in inverted
(arithmetical, logical, branch or control) MIMOPS instruction page tables, [14], stored inside the processor's fetch and decode
to be completed in а single tact cycle. The instructions that are hardware logic, whose contents are managed by the operating
supported by the proposed MIMOPS processor are MIPS alike, system. According to the implemented approach of pipelining,
but the way of their interpretation and execution is slightly MIMOPS can simultaneously access to a single instruction of
different. an on-chip instruction memory block, and to three operands of
up to three on-chip data memory blocks (some operands might
Unlike the MIPS processor that is given in Fig. 1, the be in the same block), as shown in Fig. 3.
MIMOPS processor operates directly with the on-chip memory
CPU chip
Wb_ResAddress[n-1..0] N-1
Wb_ResData[31..0]
Op2Data[31..0]
Op1Data[31..0]
10 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 3 shows how the CPU accesses to the on-chip data instructions consists of left and right logical and arithmetical
memory, during the instruction decode pipeline stage. Once the shifts and rotations. The branching group includes instructions
CPU decodes the instruction, it passes three virtual memory for conditional and unconditional change of the program flow.
addresses (for operand1, operand2, and result) to the data The last group is the auxiliary group, consisting of instructions
memory address generator unit. This unit performs a look-up in for program termination and system halt, SET instructions that
a page table in order to find the appropriate frame numbers for update the base address units, load instructions for storing 8-,
the input page numbers, and thus to generate the effective 16- or 32-bit immediate values in the on-chip data memory and
physical addresses of the two input operands and the result IN/OUT instructions for communication with external devices.
operand. After that, the physical address of operand1 and
operand2 are passed to the data memory over the memory bus, The execution of MIMOPS instructions is managed by the
while the physical address of the result operand, control signals generated by the control unit that is specifically
(Dec_ResAddress), is sent to the next CPU pipeline stage, to defined for the MEMRISC architecture. This unit provides
be further used during the write-back stage. support for several addressing modes, including base, direct,
immediate, PC-direct and PC-relative. In addition to the
According to Fig. 3, the CPU can simultaneously perform control unit, the MIMOPS processor also includes: arithmetical
two reads and a single write to the on-chip data memory. This - logical unit that can operate with integers and floating-point
is achieved in such a way that the processor fetches two 4-byte numbers, units for pipelining support, hazard detection unit for
(32-bit) data operands, starting at the generated physical overcoming data hazards during pipeline execution of
addresses of operand1 and operand2, and in parallel stores the instructions, units that provide hardware support for virtual
received 32-bit result data (Wb_ResData), starting at the memory (memory segmentation in blocks, page tables etc),
physical address (Wb_ResAddress) of the result operand, mechanisms for exception handling (ex. incorrect result), I/O
which is actually passed from the write-back pipeline stage. (in-/output) control, and additional control and selection logic.
Similarly to the result data and address forwarding
(Wb_ResData, Wb_ResAddress), the fetched operands The proposed MIMOPS processor with MEMRISC
architecture is expected to save many timing and hardware
(operand1 and operand2) are sent to the next CPU pipeline
stage, to be further used as input operands for computing some resources since it removes the complex cache memory mana-
ALU operation in the execute stage. gement mechanisms and eliminates the use of explicit load and
store instructions. Indeed, the MIMOPS processor excludes the
When it comes to pipelines, it can be noticed that both many redundant copies of data that occur in GPRs and caches
MIPS and MIMOPS processors provide overlapping of the of processors which operate with standard memory hierarchy.
execution of the instructions, by implementing pipeline This way of operation is very suitable for applications that
registers for every inter-phase (ex. instruction fetch/instruction perform many arithmetical-logical operations over some data
decode). Besides these similarities, the MIMOPS processor set that is accessed with a high degree of locality. Examples of
differs from the MIPS processor in many ways, since it allows: such type of applications are those that perform computations
reducing of the pipeline stages number by one, finishing the with matrices, such as matrix multiplication programs.
execution of conditional and unconditional branches in the
decode pipeline stage and support of data forwarding for In order to present the performance gains (in terms of
overcoming data hazards during parallel instructions execution. speed) of the novel MEMOPS processor with MEMRISC
Additionally, the MIMOPS processor implements a separate architecture, a comparative analysis between three similar
shifter logic that is purposed to generate a second flexible processors is made. This refers to a MIMOPS processor, a
operand for the arithmetical-logical unit (ALU). This is register-less PERL processor, and a RISC-based MIPS
achieved by shifting the second operand by a specific constant processor. It is considered that the proposed MIMOPS
value before it is being used by the ALU (this is similar to the processor includes on-chip memory with a capacity equal to
ARM - Advanced RISC Machine architecture, [39]). There- the amount of cache memory into the MIPS and PERL
fore, the ALU of the MIMOPS processor is able to perform processors (128KB L1 and 2M L2 cache). The actual analysis
operations over two integer or floating-point input numbers, measures the execution time of a 32x32 matrix-multiplication
where the second operand might be previously shifted. program for each of the given processors. The program
simulation is done with a MIMОPS instruction-set simulator,
Basically, the instruction set architecture of the proposed explained in [41], a MARS simulator for MIPS, [42] and a
MIMOPS processor is RISC-like and includes three types of special instruction-set simulator for PERL, given in [19].
instructions (M-type, I-type and J-type), organized in four
different groups. M-type instructions operate with memory The results of the analysis are shown in Fig. 4 and Fig. 5,
operands placed in the on-chip data memory (similar to where Fig. 4 shows the execution time of the test program run
registers in R-type MIPS instructions), while I-type and J-type on each of the three processors (PERL, MIPS, MIMOPS),
instructions operate with immediate values, whereas J-type while Fig. 5 illustrates the improvement that is achieved by
MIMOPS. Referring to these results, it can be noticed that
instructions are used for unconditional branching. Depending
on the function of the instructions, they can belong to PERL provides an improvement of 8.82% in comparison to
arithmetical-logical, shifting, branching or control group. The MIPS, but on the other hand the MIMOPS processor
arithmetical-logical group of instructions includes addition outperforms both of them, achieving 1.33 times (25%) better
with overflow detection, subtraction, multiplication, integer results than MIPS and 1.21 times (17.7%) better results than
division (div), modulo division (mod) and AND, OR, XOR and PERL. This analysis is made just to show and emphasize the
NOT logical bit-wise operations. The shifting group of performance potential of the proposed MIMOPS processor.
11 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
800000
The fetch module is purposed to read an instruction from
the on-chip instruction memory and to generate the next PC
700000 value that will be used for instruction fetch in the next tact
600000 cycle. This module includes three separate components:
500000 instruction memory, instruction page table and IF/ID pipeline
MIPS
register; that are accessed during the instruction fetching phase.
400000
PERL
300000 The decode module is purposed to decode the instruction
MIMOPS that is sent from the fetch module and to read the instruction
200000
operands that are placed inside the on-chip data memory.
100000 Besides that, this module also executes shift and sign-extension
0 operations for immediately-addressed operands, comparisons
Processor cycles for conditional branching, data hazards detection, and produces
control signals with the control unit. This module includes
Fig. 4. Execution Time of 32x32 Matrices Multiplication on Three different several separate components: data memory, data page table,
Processors: MIPS, PERL and MIMOPS.
ID/EX pipeline register, comparator, control unit, and a few
multiplexers and extenders; that are accessed during the
30 instruction decoding phase.
25 The execute module is purposed to execute shifting and
20 arithmetical-logical operations and to select the result value
(result from ALU, result from the shifter, etc.) that should be
15 compared to MIPS
written back to the on-chip data memory. In addition to that,
10 compared to PERL this module also performs forwarding of the result value and
address to the decode module in order to prevent the
5
occurrence of data hazards. This module includes several
0 separate components: ALU for integer and real numbers,
Percents (%) shifter, EX/WB pipeline register, result selector multiplexer,
and several other multiplexers; that are accessed during the
Fig. 5. Percentage Speedup of Execution Time of 32x32 Matrices instruction executing phase.
Multiplication on MIMOPS Processor.
The write-back module is purposed to write the result value
IV. FPGA IMPLEMENTATION OF THE PROPOSED RISC- to the on-chip data memory and to provide forwarding of the
BASED MEMORY-CENTRIC PROCESSOR ARCHITECTURE result to the decode module in order to prevent the occurrence
The proposed MIMOPS processor is described in VHDL, of data hazards. This module acts as an interface to the decode
by means of Xilinx VIVADO Design Suite. This software module, which actually executes the operations of writing and
environment enables hardware designers to synthesize resolving data conflicts.
(compile) their HDL codes, perform timing analysis, examine The I/O communication module is purposed to transfer data
RTL diagrams, simulate a design's reaction to different stimuli, between an I/O device and the MIMOPS processor (instruction
and configure (program) a target FPGA device. In fact, all or data on-chip memory) with IN or OUT instructions.
these functionalities are achieved by several different tools, Accordingly, this module uses an in/out data bus to receive
including: Vivado regular synthesis and XST (High-Level data from an I/O device to its on-chip memory (when IN
Synthesis) compiler, Vivado implementation tool (translate, instruction is executed) or to send data to an I/O device from its
map, place, and route), Vivado Intellectual Property integrator, on-chip memory (when OUT instruction is executed).
Vivado Simulator, Vivado serial I/O and logic analyzer for
debugging, XDC (Xilinx Design Constraints) tool for timing
constraints and entry, Vivado programming (Xilinx impact)
tool etc. In general, Vivado is a design environment for FPGA
products from Xilinx and is tightly-coupled to the architecture
of such chips. Therefore, we use the Vivado tools suite in order
to perform FPGA implementation of the proposed MIMOPS
processor on Virtex7 VC709 Xilinx evaluation platform, [32].
The VHDL model of the proposed MIMOPS processor is
organized in four modules (fetch, decode, execute, write-back)
that form the processor's pipelined data-path and an additional
module that provides communication with I/O devices. This is
also presented in Fig. 6, where a block diagram (schematic) of
the VHDL model of MIMOPS processor, generated in Vivado FE DE EX WB and I/O
Design Suite, is given. Fig. 6. Block diagram of the VHDL model of MIMOPS.
12 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Each of the given VHDL modules is represented with a analyze the processor's behavior during the execution of a test
block diagram (schematic) that is generated by the Vivado program that is placed in the processor's instruction memory,
Design Suite. In addition to that, the Vivado Simulator is used (given in Fig. 7(a)). Additionally, it is considered that the
to verify the operation of these VHDL modules with separate processor's data memory is already filled with data, as shown
test-bench programs, written for that purpose. Finally, the in Fig. 7(b). The results of the test-bench simulation are
complete MIMOPS processor is simulated, and its overall presented in Fig. 7(c).
functionality is verified. Therefore, a test-bench is written to
13 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Once the VHDL model of the MIMOPS processor is and performs management of memory blocks etc). In addition
simulated and verified, the next step is to perform synthesis to that, the MIMOPS processor includes a more complex
and implementation of the particular processor in Vivado control unit that provides support for direct access to memory
Design Suite. These activities are performed automatically operands. Besides the control unit, additional complexity is
with the synthesis and implementation tools, which are introduced with the implementation of data hazard detection
previously set to target the processor's FPGA realization on unit, comparison logic purposed for conditional branching in
Virtex7 VC709 evaluation board, shown in Fig. 8. In general, the decode phase, ALU unit that is extended to operate with
the VC709 evaluation board provides a hardware environment floating-point numbers and shifter unit that provides support
for developing and evaluating designs targeting Virtex7 for second source flexible operand. All these hardware units
XC7VX690T-2FFG1761C FPGA, [32]. This board allows are implemented with the aim to improve the computing
features common to many embedded processing systems, such performances of the MIMOPS processor, which is actually
as DDR3 memories, an 8-lane PCI Express interface, general- achieved, but the chip complexity is increased.
purpose I/O, and a UART interface. Other features can be
added by using mezzanine cards attached to the VITA-57 In order to program the Virtex7 VC709 FPGA, a constraint
FPGA mezzanine connector (FMC) provided on the board. file has to be prepared. This file is used to assign the VHDL
code signals of the MIMOPS processor to the device pins
In the synthesis stage, the VHDL model of the MIMOPS found on the Virtex7 VC709 evaluation board. For example,
processor is converted to a "netlist", which is composed of the reset signal is assigned to the on-board CPU reset push
generic circuit components interconnected with connections. button switch, which allows the user to manually reset the
After the synthesis, the Vivado implementation tool is used to processor. Similarly, the CLK signal is assigned to the 200
perform: translate, map, place, and route sub-steps. This way, MHz system clock of the FPGA board that is active on a
the MIMOPS processor is translated and mapped to Xilinx positive edge. In addition to that, the last 8 bits of the
Virtex7 XC7VX690T FPGA components and after that these ResultData signal that is forwarded from executing to the
components are physically placed and connected together write-back stage are assigned to the 8 user LEDs of the FPGA
(routed) on the appropriate FPGA board. Fig. 9 presents the board. More details about the Virtex7 VC709 board I/O pin
state of the Virtex7 VC709 FPGA device after the synthesis assignments are given in Table 1.
and implementation of the MIMOPS processor.
After the FPGA programming, the user can analyze the
Once the processor's implementation is finished, more execution of some program that is already loaded inside the
detailed reports about the hardware characteristics of the processor's on-chip instruction memory, just by observing the
designed MIMOPS processor are generated. According to the changes of the LEDs state. It is considered that the given
resource utilization report, shown in Fig. 10 it can be noticed program operates with numbers that are in the range of [0-255].
that the proposed MIMOPS processor can be implemented in Considering that a MIMOPS processor that works with 200
Virtex7 VC709 evaluation platform, by utilizing less than 1% MHz system clock executes the program very fast, an
of the slice registers and 36% of the slice LUT resources. This additional component is defined in order to scale the input 200
result is expected since the MIMOPS processor integrates the MHz clock signal to 1 Hz clock signal (with a period of 1 s).
memory inside the chip and it implements complex This way, the state of the LEDs changes slowly, so the user can
mechanisms that provide hardware support for virtual memory easily monitor the test program's execution.
(includes memory address generators with on-chip page tables
14 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
15 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
the proposed RISC-based memory-centric processor is first [8] D. Jakimovska, A. Tentov, G. Jakimovski, S. Gjorgjievska, M.Malenko,
simulated, by means of Xilinx VIVADO Design Suite “Modern processor architectures overview,” in Proc. of XVIII
International Scientific Conference on Information, Communication and
Simulator tool. The simulation is performed with test bench Energy Systems and Technologies, Bulgaria, 2012, pp. 239-242.
programs that generate timing diagrams, which are further used [9] R. Eigenmann, D. J. Lilja, “Von Neumann computers,” in Wiley
for analyzing the behavior of the hardware model of the Encyclopedia of Electrical and Electronics Engineering, Volume 23,
proposed processor and its components. The VIVADO 1998, pp. 387-400.
synthesis and implementation tools are next employed in [10] A. Bakshi, J. Gaudiot, W. Lin, M. Makhija, V. K. Prasanna, W. Ro, C.
creating an RTL model of the proposed processor and Shin, “Memory latency: to tolerate or to reduce?,” in Proc. of 12th
implementing the synthesized processor in Virtex7 VC709 Symposium on Computer Architecture and High Performance
Computing, 2000.
FPGA board. The reports that are generated from these tools
[11] S. Borkar, A. A. Chien, “The future of microprocessors,” in
present that the MIMOPS processor utilizes less than 1% of the Communications of the ACM, Vol. 54 No. 5, May 2011, pp 67-77.
slice registers and 36% of the slice LUT resources. The I/O
[12] Intel Corporation, “New microarchitecture for 4th gen. Intel core
mapping of the MIMOPS processor interfaces with the Virtex7 processor platforms,” Product Brief, 2013.
VC709 FPGA board pins and the programming of the given [13] W. Bao, S. Tavarageri, F. Ozguner, P. Sadayappan, “PWCET: power-
FPGA Virtex7 VC709 FPGA board are performed at the final aware worst case execution time analysis,” in Proc. of 43rd International
stage. The created hardware prototype is used for simulating Conference on Parallel Processing Workshops, 2014.
and analyzing of the proposed MIMOPS processor in real [14] P. Machanick, “Approaches to addressing the memory wall,” Technical
hardware, by means of Virtex7 VC709 FPGA component. This Report, School of IT and Electrical Engineering, University of
approach makes use of FPGA re-programmability, which has Queensland Brisbane, Australia, 2002.
proven to be an ideal solution for achieving reasonable speed at [15] C. Kozyrakis, D. Patterson, “Vector vs. superscalar and VLIW
architectures for embedded multimedia benchmarks,” in Proc. of the
a low price. 35th International Symposium on Microarchitecture, Instabul, Turkey,
The proposed MIMOPS processor provides many November 2002.
advantages, especially in terms of processing speed, but on the [16] J. Silc, B. Robic, T. Ungerer, Processor architecture: From Dataflow to
Superscalar and Beyond, Springer, 1999.
other hand it imposes additional requirements to the system's
[17] N. FitzRoy-Dale, “The VLIW and EPIC processor architectures,”
hardware and software, which cause limitations in its Master Thesis, New South Wales University, July 2005.
application area. Accordingly, the proposed MIMOPS
[18] M. Smotherman, “Understanding EPIC architectures and
processor implements specific ISA and several special-purpose implementations,” in Proc. of ACM Southeast Conference, 2002.
hardware components that provide direct operation with the [19] P. Suresh, “PERL - a register-less processor,” PhD Thesis, Department
on-chip memory. Therefore, it is obvious that а specific of Computer Science & Engineering, Indian Institute of Technology,
software support for the proposed MIMOPS processor should Kanpur, 2004.
be developed in the future. The primer requirement would be [20] P. R. Panda, N. D. Dutt, A. Nicolu, “On-chip vs. off-chip memory: the
designing of a dedicated compiler that would be able to data partitioning problem in embedded processor-based systems,” ACM
translate high-level language programs to MIMOPS assembler, Transactions on Design Automation of Electronic Systems, 2000.
(that significantly differs from the assembler of other RISC- [21] V. Venkataramani, M. Choon Chan, T. Mitra, “Scratchpad-memory
management for multi-threaded applications on many-core
based processors) while keeping the standard programming architectures,” ACM Transactions on Embedded Computing Systems,
model. Afterward, the next research activities would include Vol. 18, Issue 1, 2019.
developing of a dedicated operating system with process [22] C. Cojocaru, “Computational RAM: implementation and bit-parallel
scheduler, which would be able to manage the MIMOPS on- architecture,” Master Thesis, Carletorn University, Ottawa, 1995.
chip memory and to coordinate the complete virtual address [23] H. Tsubota, T. Kobayashi, “The M32R/D, a 32b RISC microprocessor
space, while multiple processes are being executed. with 16Mb embedded DRAM,” Technical Report, 1996.
Furthermore, assuming the recent innovation in processing in [24] J. Draper, J. T. Barrett, J. Sondeen, S. Mediratta, C. W. Kang, I. Kim, G.
memory architecture and technology it may become desirable Daglikoca, “A prototype processing-in-memory (PIM) chip for the data-
to build a scalable multi-processor MIMOPS-based system in intensive architecture (DIVA) system,” Journal of VLSI Signal
Processing Systems, Vol. 40, Issue 1, 2005, pp. 73-84.
very near future.
[25] M. Gokhale, B. Holmes, K. Jobst, “Processing in memory: the Terasys
REFERENCES massively parallel PIM array,” IEEE Computer Journal, 1995.
[1] D. A. Patterson, J. L. Hennessy, Computer Organization and Design: [26] K. Keeton, R. Arpaci-Dusseau, and D.A. Patterson, "IRAM and
The hardware/software Interface, 5th ed., Elsevier, 2014. SmartSIMM: overcoming the I/O bus bottleneck", in Proc. of the 24th
[2] J. L. Hennessy, D. A. Patterson, Computer Architecture: A Quantitative Annual International Symposium on Computer Architecture, June 1997.
Approach, 5th ed., Morgan Kaufmann Publishers, 2011. [27] C. E. Kozyrakis, S. Perissakis, D. Patterson, T. Andreson, K. Asanovic,
[3] “Moore's law is dead - long live Moore's law,” in IEEE Spectrum N. Cardwell, R. Fromm, J. Golbus, B. Gribstad, K. Keeton, R. Thomas,
Magazine, April 2015. N. Treuhaft, K. Yelick, “Scalable processors in the billion-transistor era:
IRAM,” IEEE Computer Journal, Vol. 30, Issue 9, pp 75-78, 1997.
[4] J. Hruska, “Forget Moore’s law: hot and slow DRAM is a major
roadblock to exascale and beyond,” in Extreme Tech Magazine, 2014. [28] J. Gebis, S. Williams, D. Patterson, C. Kozyrakis, “VIRAM1: a
mediaoriented vector processor with embedded DRAM,” 41st Design
[5] W. A. Wulf, S. A. McKee, “Hitting the memory wall: implications of Automation Student Design Contest, San Diego, CA, 2004.
the obvious,” in ACM SIGARCH Computer Architecture News, Vol.
23, Issue 1, March 1995. [29] K. Murakami, S. Shirakawa, H. Miyajima, “Parallel processing RAM
chip with 256 Mb DRAM and quad processors,” in Proc. of Solid-State
[6] Y. Yan, R. Brightwell, X. Sun, “Principles of memory-centric Circuits Conference, 1997.
programming for high performance computing,” in Proc. of Workshop
on Memory Centric Programming for HPC, USA, 2017. [30] S. Kaxiras, D. Burger, J. R. Goodman, “DataScalar: a memory-centric
approach to computing,” Journal of Systems Architecture, 1999.
[7] D. Patterson, “Latency lags bandwidth,” in Communications of the
ACM, Vol. 47, No. 10, 2004, pp 71-75.
16 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[31] M. Oskin, F. T Chong, T. Sherwood, “Active pages a computation of the 29th International Conference Architecture of Computing
model for intelligent memory,” in Proc. of the 25th Annual International Systems, Germany, 2016.
Symposium on Computer architecture, 1998, pp. 192-203. [38] E. Vermij, L. Fiorin, R. Jongerius, C. Hagleitner, J. Van Lunteren, K.
[32] Xilinx, “VC709 evaluation board for the Virtex-7 FPGA,” User Guide, Bertels, “An architecture for integrated near-data processors,” ACM
2019. Transactions on Architecture and Code Optimization, Vol. 14, Issue 3,
[33] J. M. P. Cardoso, M. Hubner, Reconfigurable Computing: From FPGAs 2017.
to Hardware/Software Codesign, Springer-Verlag New York, 2011. [39] Hewlett Packard Labs, “The machine: the future of technology,”
[34] S. Li, K. Chen, J. B. Brockman, N. P. Joupp, “Performance impacts of Technical Paper, 2016.
non-blocking caches in out-of-order processors,” Technical Paper, 2011. [40] D. Efnusheva, A. Cholakoska, A. Tentov, “A survey of different
[35] S. Ghose, K. Hsieh, A. Boroumand, R. Ausavarungnirun, O. Mutlu, approaches for overcoming the processor-memory bottleneck,”
“The processing-in-memory paradigm: mechanisms to enable adoption,” International Journal of Computer Science & Information Technology,
in book: Beyond-CMOS Technologies for Next Generation Computer Vol. 9, No. 2, April 2017.
Design, 2019. [41] G. Dokoski, D. Efnusheva, A. Tentov, M. Kalendar, “Software for
[36] G. Singh, L. Chelini, S. Corda, A. Javed Awan, S. Stuijk, R. Jordans, H. explicitly parallel memory-centric processor architecture,” in Proc. of
Corporaal, A. Boonstra, “A review of near-memory computing Third International Conference on Applied Innovations in IT, 2015.
architectures,” in Proc. of the 21st Euromicro Conference on Digital [42] K. Vollmar, P. Sanderson, “MARS: an education-oriented MIPS
System Design, 2018. assembly language simulator,” in Proc. of the 37th SIGCSE Tech.
[37] E. Azarkhish, D. Rossi, I. Loi, L. Benini, “Design and evaluation of a Symposium on Computer Science Education, 2007.
processing-in-memory architecture for the smart memory cube,” in Proc.
17 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Honeybee computing is a concept based on Authorization and authentication [8][9] are security issues
advanced ubiquitous computing technology to support Smart that must be considered during the development of an
City Smart Village (SCSV) initiatives. Advanced ubiquitous application. There are multiple cloud service providers with
computing is a computing environment that contains many client authentication method, for example Amazon Web
devices. There are two types of communication within Honeybee services that use HMAC-Sha1, HMAC-Sha256, or X.509
computing: client server and peer-to-peer. One of the certificate, Azure uses SAML 2.0 or Auth 2.0, Azure Storage
authorization techniques is the OAuth technique, where a user uses HMAC-Sha256, and Google App Engine uses OAuth 2.0,
can access an application without creating an account and can be shared secret or certificate. HTTP authentication [6][10][11]
accessed from multiple devices. OAuth is suitable to control the
provides basic and digest access authentication.
limited access of resources to the server. The server use REST
API as web service to publish data from resources. However Honeybee computing needed authorization authentication
since Honeybee computing also supports peer-to-peer that support the architecture, since the Honeybee computing
communication, security problem can still be an issue. In this support peer-to-peer and client server, secure communication
paper, we want to propose the design of a secure data during data transfer is important to protect the resource. Client
transmission for Honeybee computing by adopting the server use authorization and authentication that involved
authorization process of OAuth 2.0 and Elliptic Curve Diffie- storing of key in server side, while peer to peer security
Hellman (ECDH) with HMAC-Sha. This article will also discuss
mechanism usually involved with encryption and decryption.
the communication flow after adopting OAuth 2.0 and ECDH to
There is no security method for secure communication with
the computing environment.
both client server and peer to peer communication. This paper
Keywords—HMAC-Sha; REST API; peer-to-peer; web service; discusses the authorization and authentication process in
honeybee computing Honeybee computing.
The rest of this paper is organized as follows: The existing
I. INTRODUCTION
work of the attack, client server and peer to peer method to
Honeybee computing is a concept based on advanced secure the communication is presented in Section 2. In
ubiquitous computing technology to support Smart City Smart Section 3, we present the overview of Honeybee Computing,
Village (SCSV) initiatives1. It is supported by a middleware before discussing the findings in Section 4. Section 5 presents
together with a number of tools such as semantic knowledge the communication flow. Finally, Section 6 concludes the
tool and predictive analytics for information management. The paper and presents the future work.
sources of information in Honeybee Computing are from the
web, public and private cloud, and user devices. Since there is II. RELATED WORK
a multiple sources of data, it is important that all transactions
A. MITM Attack
are secured.
Generally, there are three types of MITM attack, namely
In the development of a software, the effort to secure the Address Resolution Protocol (ARP) Cache Poisoning, Domain
software is important, for example a framework cyber security Name System (DNS) Spoofing and Session Hijacking [12].
[1] strategy framework is to protect government data, foreign The attacks that use ARP spoofing [13] refers to a technique
investment and citizens. With many types of attacks, the that enables an attacker to pretend to be one of the users in a
importance of security is not only to look at securing the data communication between two users. The DNS spoofing
but also to ensure users authenticity [2], especially if the principle [13] is where the victim’s HTTP traffic is intercepted.
interaction involves third party users. For example, a design for The program analyzes incoming HTTPS links and replaces
a virtual private network [3] for collaboration specialist users them with unprotected HTTP links or homographic ally related
where the authentication becomes the main part of the design secure links. Session hijacking is the hijacking of a valid
and authentication mechanism for an ad-hoc network [4]. One computer session to the browser. The aim of MITM [12] is to
of the popular security problems within a network is the man- compromise the confidentiality, integrity and availability of
in-the-middle (MITM) attack [5][6]. The problem of a MITM messages. Based on the three effects, the scenario that would
attack is more critical in applications that use the single sign on be caused by MITM would be as follows:
(SSO) method. The Facebook platform that is based on cloud
computing is open to multiple types of MITM attacks [5][7].
1
GSIAC Smart City. “ Smart City-Smart Village”.
http://gsiac.org/index.cfm?&menuid=36#sthash.74DYwYR0.dpuf [28 January
2015].
18 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
19 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
3
N. Elenkov,”ECDH on Android sample app”. Github.
https://github.com/nelenkov/ecdh-kx.
20 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 5 shows that there are two ways to access the services In this system, a programmer is a user, thus user sign up is
provided by the Honeybee Security manager. A Honeybee needed and then the user must register as a programmer to
application on a PC can access the Security Web API directly. access the developer dashboard.
A Honeybee app on a mobile device will access the Security
Web API through an SDK provided by org.honeybee.security. For an end-user to develop an app, registration is needed
where the end-user needs to agree to the programmer
There are two types of services provided by the Honeybee agreement. Apps registration is needed for the Honeybee apps
manager: to access the Honeybee API. An app ID and app secret is
provided to the end-user for development purposes; the
1) Authentication and authorization services to access sequence is shown in Fig. 6.
services provided by other parts of the middleware; and
2) Encryption and decryption services for communication IV. AUTHORIZATION IN HONEYBEE COMPUTING
between apps on different devices. Honeybee computing authorization follows the OAuth
Authentication is needed to access the Honeybee workflow. The authorization is processed using multiple
middleware. A server as a security manager provides service to predefined URLs, called endpoints. There are 4 endpoints:
ensure whether the app is authorized to access sources. For 1) Request URI (this endpoint passes the request token).
user sign up, login, and logout, a link to the Honeybee main 2) Access URI (exchanges request token for an access
web page is provided. After an authentication, a user has a
token).
session ID that is stored in cookies. The rules are as follows:
3) Authorize URI (confirms that the access token is valid).
1) Programmer needs to register the apps to obtain 4) Refresh token (refresh access token if previous is
information such as apps id and apps secret. invalid).
2) End_user is granted the type of permissions for the apps
Fig. 7 shows the sequence diagram between user,
to access the source. programmer, client, and server. The programmer develops the
3) Each device provides a MAC address to access the Honeybee app or web application and interacts with the
server, and every token is provided for each device at each Honeybee client before redirecting to the endpoints in the
request to the server. server. This process is adapted from the OAuth2.0 security.
Each number in the diagram is explained as follows:
Request URI: The first endpoint is the request URI.
This request URI is provided by the packages in
Honeybee computing. The request URI passes the app
secret to check if the user has granted permission to the
Honeybee app to access the user account.
Access URI: In the second endpoint, after the user is
granted permission, an access token is provided. This
access token can be used for requests to the resource
server.
Authorize URI: In this third endpoint, the access token
is checked whether it is valid or not; if the token is valid
Fig. 5. Secret Generated by HMAC-SHA1 Algorithm for Programmer to
Develop Application. then data is returned from the resource server.
Refresh token URI: The fourth endpoint is used to
refresh expired tokens; this is used when the user opens
the app and the token saved is expired. This endpoint is
not shown in the diagram. From Fig. 7, users must grant
permission to enable third party applications to the
access server.
The Honeybee computing Security manager is responsible
for request management. This is to ensure that all requests are
authorized. This is because every resource in Honeybee
computing needs permission from the user and tokens to get
authorized. The template is designed so that author affiliations
are not repeated each time for multiple authors of the same
affiliation. Please keep your affiliations as succinct as possible
(for example, do not differentiate among departments of the
Fig. 6. Sequence Diagram in Security Manager. same organization). This template was designed for two
affiliations.
21 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
22 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
23 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—A method for agricultural product quality and NDVI estimation and its application for insect damage
harvest amount prediction by using smartphone camera image is detection of rice paddy fields is discussed [6]. Artificial
proposed. It is desired to predict agricultural product quality and Intelligence: AI baaed fertilizer control for improvement of rice
harvest amount as soon as possible after the sowing. In order for quality and harvest amount is proposed and well validated the
that, satellite imagery data, UAV camera based images, ground proposed method with rice paddy field data. [7]. Method for
based camera images are used and tried These methods do cost NIR reflectance estimation with visible camera data based on
significantly and these do not work so well due to some reasons, regression for NDVI estimation and its application for insect
in particular, most of farmers cannot use these properly. The damage detection of rice paddy fields is proposed [8]. Also,
proposed method uses just smartphone camera acquired images.
artificial intelligence baaed fertilizer control for improvement
Therefore, it is totally easy to use. If the results of prediction of
product quality and harvest amount are not satisfied, then
of rice quality and harvest amount is proposed [9].
farmers have to add some additional fertilizer at the appropriate It is not always possible to acquire the remote sensing
time. The experimental results with soy plantations show some satellite data due to the limitation of revisit cycle of the satellite
possibility of the proposed method. orbit. Neither, it is not possible to acquire UAV camera images
due to the limitation of wind speed, weather condition, and so
Keywords—Smartphone camera image; agricultural product on. On the other hand, ground based cameras have problems on
quality and harvest prediction; fertilizer control; soy plantation
sun illumination condition difference, shading and shadowing,
I. INTRODUCTION etc. Moreover, these methods do cost very much.
Isoflavones are most specific feature of soy. Therefore, soy Meanwhile, the proposed requires only smartphone camera
farmers intend to find species which is isoflavone rich species derived images. Therefore, it is easy to use and does not
and also intend to control water supply, fertilizer etc. for require much cost. Only thing the farmers have to do is just
improving soy quality and harvest amount. Because acquire images of the soy plantations (example of agricultural
isoflavones are ubiquitous in the germ part. Therefore, it is not products) and send these to the image collection center. After
so easy to estimate the isoflavone content in the planted soy. the farmers send their acquired images, the image collection
center analyzed data, and predict their product quality and
Nitrogen content estimation of rice crop based on Near harvest amount and send the predicted result with some
Infrared (NIR) reflectance using artificial neural network appropriate instructions for fertilizer control (fertilizer amount
(ANN) is conducted [1]. Rice crop field monitoring system and timing) to the farmers.
with radio controlled helicopter based near infrared cameras
through nitrogen content estimation and its distribution In the following section, the proposed prediction method
monitoring is proposed [2]. Rice crop quality evaluation and system will described followed by experimental set-up
method through regressive analysis between nitrogen content together with experimental results. After that, concluding
and near infrared reflectance of rice leaves measured from near remarks and some discussions will be described.
field radio controlled helicopter is also proposed and well
reported [3]. Furthermore, a relation between rice crop quality II. PROPOSED METHOD
(protein content) and fertilizer amount as well as rice stump A. System Configuration
density derived from helicopter data is well reported [4]. Then
Fig. 1 shows the proposed system configuration and the
estimation of rice crop quality and harvest amount from
procedure. Following is the detailed procedure:
helicopter mounted NIR camera data and remote sensing
satellite data is proposed [5]. 1) Farmer sends smartphone images to the image
Effect of stump density, fertilizer on rice crop quality and collection center.
harvest amount in 2015 investigated with drone mounted NIR • Register the system
camera data is also confirmed [6]. Method for NIR reflectance • Report conditions
estimation with visible camera data based on regression for
24 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
III. EXPERIMENTS TABLE. II. MEAN, STANDARD DEVIATION, MINIMUM AND MAXIMUM
PIXEL VALUES OF THE RECTANGLE AREAS
A. Experiment Procedure
June 4 S June 27 S July 12 S July 12 F
Sowing time, fertilizer time and flowing time of the two
types of soy (Sakukei 207: New specie, and the traditional Mean 125.48 135.1 106.11 110.85
Fukuyutaka) are shown in Table I. Smartphone camera data are
Std. 19.25 20.35 35.16 38.38
acquired on the following dates:
Min. 60 58 38 22
July 26, August 17, 22, 27, September 11, 25 in 2018
Max. 218 225 230 223
25 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
(a) Test Site on Map (Test Site is Situated at the Red Circle) (b) SPARI (c) Test Site on 3D Aerial Photo Image of Google Map.
Fig. 2. Test Site.
(a)Sakukei207(6/4) (b)Sakukei207(6/27)
26 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
ref800
Meanwhile, spectral reflectance of the soy plantations is 0.4
measured. Therefore, NIR image (at around 800nm) of image 0.3
can be estimated with visible colored smartphone camera 0.2
images. Example of the measured spectral reflectance is shown 0.1
in Fig. 4. 0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
If just reflectance at 800nm of the soy plantation measured number
horizontally is taken into account, then Fig. 5 of horizontal
Fig. 5. Horizontal Profile of the Measured Reflectance of Soy Plantation.
profile of the measured reflectance of soy plantation can be
shown. y = 15.832x + 32.883
50
Harvest R² = 0.3614
These reflectance are measured on September 25 2018.
Harvest amount*0.1(kg/10a),
Protein content(%), Water
40 Protein
E. Measured Truth Data
The measured truth data are as follows: Water
content(%)
30
Soy isoflavone, Phospholipid, Nitrogen content, water Linear y = 61.857x - 6.0371
content and protein content in the harvested soy beans and 20 (Harvest) R² = 0.3501
harvest amount. Firstly, the following correlation analysis is Linear
conducted with the measured reflectance of the soy plantation 10 (Protein)
and protein content, water content, as well as harvest amount. Linear y = -7.2315x + 19.63
(Water) R² = 0.1102
F. Correlation Analysis and Linear Regression Between 0
Smartphone Camera Derived Reflectance at 800nm and 0.40 0.45 0.50 0.55 0.60 0.65 0.70
Ref800
the Truth Data
Fig. 6. The Relation between the Measured Reflectance of Soy Plantation at
The relation between the measured reflectance of soy 800nm and Water Content, Protein Content of the Harvested Soy Beans and
plantation at 800 nm and water content, protein content of the Harvest Amount.
harvested soy beans and harvest amount is shown in Fig. 6.
From the figure, it is found that there is not so small correlation The regressive equations and R square values for protein
between the measured reflectance and protein content and also content in soy beans, harvest amount and water content in soy
harvest amount. Therefore, it is possible to predict these two beans are as follows:
parameters (harvest amount and protein content) from
reflectance measurement with not only spectral-radiometer, but p = 15.832x + 32.883
also smartphone camera. Even if the spectral coverage of the R² = 0.3614 (1)
smartphone camera rages from blue to red, it is still possible to
estimate the reflectance at 800nm with visible smartphone h = 61.857x - 6.0371
camera if a calibration between visible camera and spectral-
R² = 0.3501 (2)
radiometer is conducted. On the other hand, there is no such
correlation between the measured reflectance at 800nm and w = -7.2315x + 19.63
water content in the harvested soy beans.
R² = 0.1102 (3)
1.0 Therefore, it can be said that protein content in soy beans
0.9 and harvest amount can be predicted with the measured
0.8 reflectance of the soy leaves at 800 nm in some sense (around
0.6 of correlation coefficient between both). Also, it is not
0.7
possible to estimate water content in soy beans with the
0.6 measured reflectance of the soy leaves at 800 nm.
Ref800
0.5
It is also found that there is strong correlation between
0.4 maximum pixel value and soy isoflavone and Phospholipids
0.3 nevertheless the correlation between reflectance at 800nm and
0.2 soy isoflavone and Phospholipids is week. The correlation
0.1 between soy isoflavone is much stronger (0.821) than that
between Phospholipids (0.309) as shown in Table III.
0.0
350 550 750 950 Because the pixel values in the smartphone camera image
Wavelength(nm) are variated due to the angle of the soy plant leaves and camera
looking angle are so different, the maximum pixel value is
Fig. 4. Example of the Measured Spectral Reflectance of the Soy Plantations. much more appropriate for the correlation analysis.
27 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. III. CORRELATION BETWEEN MAXIMUM PIXEL VALUE OF THE SMARTPHONE CAMERA IMAGE AND SOY IS OF LAVONE AND PHOSPHOLIPIDS
28 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
29 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In this work, a safe travelling technique was kind of death or injuries caused to the animals. However, this
proposed and implemented a LoRa based application to avoid was misused by some of the camel owners to earn money by
the collision of animals with vehicles on the highways of Saudi pushing their animals on roads as reported by Ansari et al. [5].
Arabia. For the last few decades, it has been a great challenge for This has become a serious issue for the governments and the
the authorities to secure the life of animals and human being on commuters to overcome as it is created the problem. Similar
the roads due to the sudden passage of animals on the highways. problems of a motor vehicle collision with Kangaroo’s were
In such situations, drivers are not aware of the animal reported in Australia in the night times by Abu-Zidan [6].
movement, and serious damage is observed with the life of both Surprisingly the rate of accidents in winter due to the
humans and animals. A LoRaWAN based architecture with a
kangaroo-vehicle collision was reported to be very low in
variety of advantages towards low cost and high accuracy of
finding the movement of animals is possible with the proposed
Australia. The vehicle collisions with the larger animals
method and could deliver good results as well. The accuracy of generally cause severe trauma to occupants due to fatal attacks
this method was improved to a maximum extent as compared to caused by the wildlife animals [7]. Therefore, to address these
the existing system due to the usage of LoRa sensors implanted in kinds of animal-vehicle collision (AVC) or camel-vehicle
the animal’s skin to trace with the nodes and base stations easily. collision (CVC) problems various countermeasures have been
introduced in the past by many researchers with the help of the
Keywords—LoRa; Sensor-based mobile applications; runtime government in Saudi Arabia [8]. Similarly, Bashir and Abu-
monitoring; tracking; global positioning system Zidan proposed different types of preventing techniques to
avoid motor vehicle collision (MVC) with large animals [9].
I. INTRODUCTION This includes alarming signs, underpasses or overpasses for
The Middle East is facing a serious problem with the animals, frightening reflectors, etc.
presence of camels on the roads/highways and is the reason A lot of data was collected by Huijser to study different
for many accidents with an average of about 600-700 deaths in types of reasons for AVC revealed that sex and age of the
a year. Presence of the stray animals is not only the problem animals are also playing a vital role in the AVC. Sometimes
of the Middle East but also found to be a major concern in the animals become more protective and aggressive to save
world-wide countries. In daylight somehow the people are still their family or even to entertain themselves [10]. Later,
managing to escape, but in night times it is always a Huijser and Kociolek presented a detailed summary with the
challenging risk to address carefully to save the animal-human highlights related to issues, problems, and experiences of
life’s and avoids the major damages. In this context, a lot of different operations such as false positives, false negatives,
research was carried out by different authors to find maintenance, etc. [11]. It was also noted that most of the
reasonable solutions. In the early 70’s, Al-Ghamdi highlighted accidents found to be with the vehicle speeds of above 88
the 30 times rise in road accidents in the Kingdom of Saudi km/h (55mi/h) and proposed to alert the drivers to reduce
Arabia (KSA) [1]. Later in 80’s, it was reported by Tamimi et speed with a distance of almost 21 m (68 ft). A lot of research
al., that most of the accidents are due to driver errors (around has been carried out in this work towards wildlife-vehicle
90%) and they are between the ages of 20 and 40 [2]. Later, collision (WVC) and provided different types of mitigation
Qayed reported a huge number of accidents (around 6,117) in measures to prevent accidents due to animals. The author also
Saudi Arabia from June 1994 to 1995 and among which 2,551 discussed different types of animal detection systems and
people got injuries and 151 lost their lives [3]. The author in wildlife crossing structures that influence towards reducing
his work explained that due to vehicle collision with other the number of accidents in his work [12]. Similar work was
vehicle and fixed objects is mostly seen. However, the carried out by Al-Shimemeri and Arabi on Arabian camels
collision of vehicles with animals in the night times proved to and highlighted the types of injuries that occur due to the
be a critical issue to be considered carefully on the highways. collision with the Arabian camels with a weight of over 726
In a statistical analysis carried out by Ansari et al. revealed kg [13]. Many camels tend to sit and relax on the highways in
that 564,762 people (which is equivalent of 3.5% of the total the night times and are killed by accident or by some of the
population in KSA) were died or injured in road accidents [4]. ruthless drivers as shown in Fig. 1.
Later, the government of KSA carried out stringent action
against the rash driving citizens and even penalized for any
30 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 1. Shows the Huge loss of Camels due to Accidents with Vehicles on
the Highways/Roads.
31 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
32 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
a rectangle shaped vehicle passes the alert will not be given Warning Whistles: They are independent of any kind
but when the shape of the animal (non-rectangle) is identified of installations, such as the roadside equipment, etc.
means an alert will be given to the drivers with the help of This concept was introduced in the late 70’s [24]. The
broken lights. However, in this method, if a deer stands warning whistles are of air activated type and will be
between the laser beams for a longer time (more than a mounted on the front side of the vehicles. These
minute) means the warning signals will be turned OFF. The whistles produce ultrasonic frequency waves for any
biggest drawback of this method is to use only for the shortest kind of wind rushing towards them. The sounds made
distance and requires high maintenance cost. Finally, by these whistles scare away the animals and the
microwave technology using a series of transmitters and animals are supposed to run away from the roadside.
receivers animal detection is possible using continuous But it is reported that the audio signals will make any
microwave RF signals [27]. Using these method animals are effect on the behaviour of the animals [10].
alerted when a vehicle is detected passing through the road by
using the variety of noise or light signals. But this system is Infrared Detection System (IDS): In this case, the
not useful for high vehicle density roads because the noise will drivers are alerted with the help of infrared sensors
be ringing continuously with the passage of the vehicles [15]. connected to the vehicle when the animals are
identified within a range of distances. However, the
B. Animal-based Technologies range of these sensors must be within the reach of
Different types of collars are introduced in this method to animals and must be sufficient enough to allow the
mitigate the AVC by providing a blinking signal system. They drivers to apply the brakes within the time [25]. In this
are specially classified into two types: a) reflective collars and method, a monochromatic display will be used to see
b) radio collars. These two collars require a minimum the hot objects in white colour and cold objects as
infrastructure requirement and are easily available in the black in the images. This method helped the drivers to
market. However, based on the GPS systems also the collars some extent but some people reported to have
were installed on the animal’s neck or at other body parts. headaches. Apart from these, there are many
These systems can cover a long range of distance and a maintenance and technical problems while using IDS.
massive range of animals can be covered using this system Also, false positive and false negative problems as
[15]. discussed earlier are common to find by using this
method [26].
Reflective Collars: There are many methods to detect
the animals passing through the roads using sensing III. PROPOSED ARCHITECTURE
devices and technology-based devices/equipment. At present few countries have got their own GPS satellites
There are special categories of vehicle detection includes the US, Russia, European Countries, China and India.
systems available in the market, which help the drivers Therefore it will be a great deal to adopt the GPS systems in
to identify the animals from a long distance due to the its direct form as a complete solution to resolve the problem in
reflective tape on the neck of animals. But these collars Saudi Arabia due to a huge investment. Therefore in this
need to be maintained properly by the owners of the work, a LoRa based approach is proposed to deal with the
animals so that they are visible from distance and collision between animals and vehicles in the proposed system
avoidance of collision is sometimes not possible if the with the help of GPS based Google Maps and sensor-based
distance is very high [28]. mobile technology to initiate the alerts.
Radio Collars: The first time they were introduced in In the recent times, there are so many advanced
the year 1999 in Olympic Peninsula, Washington. The technologies evolved into the market that is very much
animals were radio-collared and the receivers used to suitable to use for implementing the safe travelling methods
scan the frequencies of different radio collars for the for the commuters in Saudi Arabia by avoiding the collision
whole day. In the presence of a signal at a particular with animals. There are seven technologies in the market that
radio-collar within a distance of 400 m of the roadside, are helpful to track the location. They are Beacons, Wi-Fi,
a signal activates the flashing beacons. However, due Radio-Frequency Identification (RFID), Near-Field
to the operation of these radio collars is for 24/7, the Communication (NFC), Quick Response Codes (QR), LoRa
batteries run out after several years and must be and GPS as listed in Table I with different technological
replaced, which is a serious concern using these radio parameters. The range, cost, accuracy, and security of these
collars [15]. technologies play a key role to define the type of application
GPS Collars: Movement of a large number of animals where they can be implemented with low risk. There is
can be traced out by this method and can be helpful in nothing like a winner or a loser for the tracking game out of
collecting a massive amount of the data to analyze the all the following tracking technologies [29]. In recent times,
tendency of the animal movement in different the accuracy of GPS proved to be increasing and Bluetooth
scenarios and meteorological conditions. beacons are well equipped to transfer the larger data. Above
and all the usage of mobile phones increased drastically with
C. Vehicle-based Technologies the increased competition between mobile manufacturers.
In the early stages, the technology was used in two ways to Most of these technologies proved to be excellent for tracking
avoid AVC by using: a) the warning whistles, and b) the in- the people in indoor locations.
built infrared detection system.
33 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. CLASSIFICATION OF DIFFERENT LOCATION TRACKING TECHNOLOGIES WHICH CAN BE USED FOR SAFE TRAVEL USAGE
34 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
analysis, etc. However, for the current problem, one needs to of messages and requirements of public network operators to
have a customized device, which is suitable to tag with the serve huge markets.
animals travelling in extreme weather conditions as well. The biggest advantage of LoRa lies in its ability to provide
Some of the sensors existing in the mobile technology may not efficient, flexible and with the reasonable economic solutions
be very much suitable for extreme heat conditions of Saudi for most of the real-time problems in most of the rural and
Arabia. Therefore, the selection of the devices and sensors indoor applications, where most of the cellular and WiFi
must have a careful look before they are adopted in this networks proved to be ineffective. It is a good choice for long
problem statement. In the current architecture, the sensors are range and Internet of Things (IoT) solutions with low power
being used to generate an audio alert for the drivers when the applications. It also enables different types of tracking
movement of animals found to be communicated by the LoRa applications in the absence of GPS and the LoRaWAN
Base Station. This audio voice will be loud enough and will be protocol helps to create the interoperability among different
OFF only when the driver sitting in the car press the message applications, IoT devices, and telecommunication operators.
acknowledge button. The architecture of a LoRaWAN consists of four major
sections: a) end nodes, b) concentrator/gateway, c) network
d) LoRa Wireless RF Technology: This is also known as server and d) Application server as shown in Fig. 5.
a LoRaWAN, which is a low power wide area network
(LPWAN) grabbing the attention of a large number of people Functioning and Security of LoRa Architecture
across the globe. This is one of the latest technologies Most of the IoT devices using the LoRa consist of a
connecting the devices in rural areas and urban areas for up to combination of the sensors with LoRa transceivers at the
30 miles in indoor environments with low energy consumption endpoints. These endpoints are connected in a star network
[35]. The battery life of these devices can be up to 10 years; and transmit the signals received from the sensors when they
hence, reducing the battery replacement cost gives an edge for detect the movement of the devices consisting of LoRa
most of the tracking systems. This is one of the GPS-free sensors. The base station receives the information and passes
tracking application, which provides low power benefits as the same through different gateways using standard IP
compared to other technologies in the market. This technology connections. The data use different types of standard protocols
can be implemented as the greatest tool for the animal tracking to connect with telecom networks. The network servers
purpose due to the flexibility and adaptability with the devices manage the data based on the functions to eliminate the
in motion with low power consumption, reduced cost, duplicated packets. The application servers will control the
minimum infrastructure, battery replacements and low overall actions of the endpoints and/or collect data from the
operating expenses. It supports mobile technologies, millions endpoints.
Runtime
Microchip Animals/Camels Monitoring
Implantations Database System/
Checker
Fig. 4. Shows the Proposed Architecture for Safe Travelling in Saudi Arabia.
35 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In the LoRa messaging, the messages are transmitted by be an alert to the drivers on their mobile application with a
using the LoRaWAN network by the end devices, which helps message and caution sound. However, if the animals are
to improve the resilience of the network. Multiple base moving in the red zone (i.e. indicating a need for serious
stations in an area may increase the deployment capital but attention by the drivers) means there will be a message alert
also helps to enhance the performance. Multiple receptions are and the sensors of the mobile phone will activate a vibration
filtered at LoRa server, which also helps in providing the with ring tone to alert the drivers to be more careful and
security checking. The security on LoRa network is provided reduce the speed of the vehicle. The sequence diagram with
by unique network key (EU164) at network levels, unique important components and their interaction with different
application key (EU164) at the application level and device- components at different levels in the proposed architecture are
specific key (EU128) for the device levels.
shown in Fig. 8.
IV. FUNCTIONING OF THE PROPOSED ARCHITECTURE
The following steps will be implemented to monitor the
movement of camels on the highways/roads of Saudi Arabia.
a) Installation of LoRa Base Stations to establish the
communication between animals and Nodes (sub-nodes too):
In this work, the animals are implanted with LoRa sensors into
their skin in the first stage. On the other hand star topology-
based networks are created with nodes and sub-nodes of
specific identification techniques also. Throughout the area,
these nodes and sub-nodes are covered using the LoRaWAN.
All these nodes and sub-nodes are connected with certain base
stations. Such a combination of nodes, sub-nodes and base
stations are repeated throughout the highways at different
places as a Unit. The nodes will communicate with the base
stations when an animal (which is implanted with LoRa
sensor) is identified within the different ranges as shown in
Fig. 6. The range of a Unit will be between 10-300 m from the Fig. 6. Defined the different Zones across the Highways for Detecting the
highway divided into three zones: red zone (0-10 m), yellow Animals with Respect to the Distance between the Road and Animals.
zone (10-100 m) and green zone (100-300 m).
b) Establishing communication between the base
stations and GPS based mobile phones: The Units are
connected with the GPS systems simultaneously giving
specific information about each node to update the latitude,
longitude, altitude, etc. on a regular basis. Google maps APK
is installed with the present application which displays the
sensor data received from the animals at different locations.
The runtime monitoring system will check the database for
animal movements from the information received from the
base stations (see Fig. 7). The Checker component will try to
assess the information obtained from the base stations and
identify the animal movements in the different zones.
c) Establishing an alert system using the proposed
mobile application based on the distance between the animals
and main roads: If the animals are within the green zone (i.e.
safe zone) means there will not be any kind of alert. But if the
animal movement is in the yellow zone (i.e. the possibility of
animals rushing on to the highway is more) means there will Fig. 7. Shows the Functioning of the Network in the Proposed Architecture.
36 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
V. CONCLUSIONS REFERENCES
[1] S. Al-Ghamdi. "Road accidents in Saudi Arabia: a comparative and
In the proposed method, the usage of GPS was limited by analytical study." WIT Transactions on the Built Environment 26
using LoRaWAN to establish communication between (1970).
different nodes and sub-nodes after detecting the animals near [2] T. M. Tamimi, M. Daly, M. A. Bhatty, and A. H. M. Lufti. "Causes and
the highways/roads. The implanted LoRa sensors are capable types of road injuries in Asir Province, Saudi Arabia, 19751977:
preliminary study." Saudi medical journal 1, no. 5 (1980): 249-256.
of working for more than 10-25 years as they are not [3] M. H. Qayed. "Epidemiology of road traffic accidents in Al-Ahssaa
dependent on any of the external sources. Google Maps Governorate, Saudi Arabia." (1998).
platform was used along with the newly developed APK to [4] S. Ansari, F. Akhdar, M. Mandoorah, and K. Moutaery. "Causes and
monitor the sensor data on the mobile for any kind of animal effects of road traffic accidents in Saudi Arabia." Public health 114, no.
movement near or on the highways. The information of the 1 (2000): 37-39.
[5] S. A. Ansari, M. Mandoorah, M. Abdalrahim, and K. R. Al Moutaery.
animal movement will be transmitted to the nearest base "Dorsal spine injuries in Saudi Arabia—an unusual cause." Surgical
stations and the data will be analyzed for false alarms and neurology 56, no. 3 (2001): 181-184.
false negatives in the process in the runtime monitoring [6] F. M. Abu-Zidan, K. A. Parmar, and S. Rao. "Kangaroo-related motor
system. The location of the nodes detecting the animals will vehicle collisions." Journal of Trauma and Acute Care Surgery 53, no. 2
be regularly updated with the GPS system and then based on (2002): 360-363.
[7] T. P. Pynn, and B. R. Pynn. "Moose and other large animal wildlife
the checker component analysis the zone wise information vehicle collisions: implications for prevention and emergency
will be alerted in the form of LoRaWAN messaging services. care." Journal of Emergency Nursing 30, no. 6 (2004): 542-547.
The sensor-based mobile applications will be generating the [8] A. S. Al-Ghamdi, and S. A. AlGadhi. "Warning signs as
alerts based on the zones as defined in the above discussions. countermeasures to camel–vehicle collisions in Saudi Arabia." Accident
Analysis & Prevention 36, no. 5 (2004): 749-760.
The efficiency of the proposed method revealed satisfactory [9] M. O. Bashir, and F. M. Abu-Zidan. "Motor vehicle collisions with large
results by improving the alert quality and response as well. animals." Saudi medical journal 27, no. 8 (2006): 1116-1120.
The overall system implementation is very much cheaper as [10] M. P. Huijser. Animal-vehicle collision data collection. Vol. 370.
compared with the GPS based systems for the conditions of Transportation Research Board, 2007.
Saudi Arabia without its own GPS system. [11] M. P. Huijser, and A. V. Kociolek. "Wildlife-vehicle collision and
crossing mitigation measures: a literature review for Blaine County,
Idaho." Western Transportation Institute, Montana State University,
ACKNOWLEDGMENT Bozeman (2008).
The author would like to thank first the Almighty, Allah [12] M. P. Huijser, K. J. Paul, and L. Louise. Wildlife-Vehicle Collision and
for his grace and blessings. The author also thanks his parents, Crossing Mitigation Measures: A Literature Review for Parks Canada,
Kootenay National Park. No. 4W1929 A. Western Transportation
family and teachers for their unconditional support and Institute, College of Engineering, Montana State University, 2008.
encouragement throughout the career.
37 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[13] A. Al Shimemeri, and Y. Arabi. "A review of large animal vehicle [26] M. P. Huijser, P. T. McGowen, and W. Camel. Animal vehicle crash
accidents with special focus on Arabian camels." Journal of Emergency mitigation using advanced technology phase I: review, design, and
Medicine, Trauma and Acute Care 2012, no. 1 (2012): 21. implementation. No. FHWA-OR-TPF-07-01. Western Transportation
[14] M. S. Zahrani, K. Ragab, and A. U. Haque. "Design of gps-based system Institute, 2006.
to avoid camel-vehicle collisions: A." Asian J Appl Sci 4, no. 4 (2011): [27] K. Finkenzeller. RFID handbook: fundamentals and applications in
362-377. contactless smart cards, radio frequency identification and near-field
[15] K. Ragab. "Simulating camel-vehicle accidents avoidance communication. John Wiley & Sons, 2010.
system." International Journal of Future Generation Communication and [28] M. M. Hurwitz. "Interchangeable attachments for collars, leashes, belts
Networking 4, no. 4 (2011): 43-56. and accessories." U.S. Patent 8,142,053, issued March 27, 2012.
[16] H. M. Hassan, L. Dimitriou, M. A. Abdel-Aty, and A. S. Al- [29] Lighthouse. 2019. 6 Technologies that can be used to track location.
Ghamdi. Analysis of Risk Factors Affecting Size and Severity of Traffic [Online] available on URL: <https://blog.lighthouse.io/6-technologies-
Crashes in Riyadh, Saudi Arabia. No. 13-2333. 2013. that-can-be-used-to-track-location/>, [June 21, 2019].
[17] R. Al-Awaal. "KSA tops world’s mobile phone users." Saudi [30] Y. Grauer. A practical guide to microchip implants. [Online] available
Gazette (2014). on URL: <https://arstechnica.com/features/2018/01/a-practical-guide-to-
[18] A. A. Abdo, and A. A. Al-Ojaili. "Assessment of awareness of microchip-implants/>, [June 23, 2019].
livestock-vehicle traffic accidents in Dhofar region, [31] A. Weissbrod, A. Shapiro, G. Vasserman, L. Edry, M. Dayan, A.
Oman." International Journal of, Applied Engineering Research Yitzhaky, L. Hertzberg, O. Feinerman, and T. Kimchi. "Automated
(IJAER) 10, no. 18 (2015): 38955-38959. long-term tracking and social behavioural phenotyping of animal
[19] E. DeNicola, O. S. Aburizaize, A. Siddique, H. Khwaja, and D. O. colonies within a semi-natural environment." Nature communications 4
Carpenter. "Road traffic injury as a major public health issue in the (2013): 2018.
Kingdom of Saudi Arabia: A Review." Frontiers in public health 4 [32] M. G. T. Mowers. "Tracking device for pets." U.S. Patent 5,850,196,
(2016): 215. issued December 15, 1998.
[20] M. Khalilikhah and K. Heaslip. "Improvement of the performance of [33] P. Bertagna. How does a GPS tracking system work? [Online] available
animal crossing warning signs." Journal of safety research 62 (2017): 1- at URL: <https://www.eetimes.com/document.asp?doc_id=1278363#>,
12. [June 25, 2019].
[21] A. Ward. Lorin. Mule deer behavior in relation to fencing and [34] S. Plangi, A. Hadachi, A. Lind, and A. Bensrhair. "Real-Time Vehicles
underpasses on Interstate 80 in Wyoming. No. 859. 1982. Tracking Based on Mobile Multi-Sensor Fusion." IEEE Sensors
[22] S. M. R. Hosseini, D. Khorasani-Zavareh, and A. Abbasi. "Challenges Journal 18, no. 24 (2018): 10077-10084.
and strategies for preventing vehicle collisions with camels in South [35] SEMTECH. What is LoRa? [Online] available at URL:
Khorasan Province: a qualitative study." Safety Promotion and Injury <https://www.semtech.com/lora/what-is-lora>, [June 21, 2019].
Prevention 6, no. 1 (2018): 43-48.
[23] W. Saad, and A. Alsayyari. "Loose Animal-Vehicle Accidents AUTHOR’S PROFILE
Mitigation: Vision and Challenges." In 2019 International Conference Amr Jadi is an Associate Professor of Software
on Innovative Trends in Computer Engineering (ITCE), pp. 359-364. Engineering at Collage of Computer Science and
IEEE, 2019. Engineering, University of Hail.
[24] K. K. Knapp. Deer-vehicle crash countermeasure toolbox: a decision
and choice resource. Midwest Regional University Transportation Dr. Jadi received PhD degrees from De Montfort
Center, Deer-Vehicle Crash Information Clearinghouse, University of University and Master’s Degree from Bradford University,
Wisconsin-Madison, 2004. UK. The author is specialized in with an area interest in
[25] H. Bender. Deterrence of kangaroos from roadways using ultrasonic Software Engineering, Early warning systems, Risk
frequencies-efficacy of the Shu Roo. University of Melbourne, management and Critical Systems. Presently the author is also involved in
Department of Zoology, 2001. various development activities within the University of Hail and abroad as a
consultant.
38 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This paper presents a compact broadband antenna proposed. The different shapes of radiating patch and
for civil and military wireless communication applications. Two modified ground plane were reported in [4] and [5]. Further,
prototypes of the antenna are designed and simulated. The researchers have been proposed different techniques i.e. meta-
proposed antenna is etched on low cost substrate material with material resonators, different shapes of slots, tuning stubs and
compact electrical dimensions of 0.207λ×0.127λ×0.0094λmm3 at proper selection of the feed line [6][7][8][9]. These techniques
2GHz frequency. The simple microstrip feeding technique and were used for the improvement in impedance BW. The
antenna dimensions are involved in the design to attain the existing study proposed the compact elliptical patch-based
proper impedance matching. An optimization of variables is planar monopole antenna by embedding arc-shaped slot for
carried out by multiple rigorous simulations. The designed
UWB applications [10]. The Daisong Zhang and Yahya
antennas have achieved the broadband impedance bandwidth of
89.3% and 100% at 10dB return loss. The antennas exhibit omni
Rahmat Samii have been designed the antenna with top cross
directional radiation pattern at lower resonances and strong loop engraved on the compact substrate 0.345λ×0.575λ×0.02λ
surface current distribution across the radiator. The peak mm3 at 3.43 GHz resonant frequency. The proposed antenna
realized gain of 5.2dBi at 10.9GHz resonant frequency is realized. has achieved 91% of fractional impedance BW [11]. Another
Results reveal that the proposed broadband antenna is a better work was presented on the broadband antenna with parastic
choice for WiMAX, UWB, land, naval and airborne radar patch technique. The antenna exhibited the relative BW of
applications. 80% with compact dimensions of 0.521λ×0.521λ×0.012λmm3
particularly at 2.32GHz operating frequency [12]. However,
Keywords—Compact antenna; broadband; microstrip feeding; the designed antennas have the complex structures and larger
civil and military; peak realized gain and impedance bandwidth dimensions. Moreover, Arash Valizade et al. demonstrated the
protrude ground plane structures [13]. Jian-Feng Li et al.
I. INTRODUCTION presented the idea on the isolation of antennas with T-Shaped
In modern communication systems, the requirement of slits in the antenna design structure [14]. Asghar Mousazadeh
smart antennas is growing rapidly in the market. These et al. presented the work on the broadband antenna with
antennas are economical, small in size, light weight with inverted L-shaped grounded strips. The antenna has the
enhanced charactestics [1]. Recently, microstrip patch compact dimensions of 0.601λ×0.601λ×0.008λmm3 [15]. The
antennas with different shapes are good choice for different defected ground structure concept was utilized in [16] and
wireless communication systems. The narrower impedance [17]. Different broadband antennas were suggested in
bandwidth (BW) and larger physical and electrical dimensions [18][19][20]. However, the employed techniques focused in
are major concerns of patch antennas [2]. Therefore, antenna the literature were complex and excessive variables utilized in
design engineers are working on the enhancement of the designed antennas might result in computational
important parameters of the compact patch antennas such as complexity. In the modern antenna topology, the compact size
impedance BW, gain, stable radiation pattern and radiation and adjustment in the designed antenna dimension in terms of
efficiency. variables is required. This adjustment can be achieved by
electromagnetic (EM) simulation software which has the
Nowadays, researchers are paying attention on the design capability of the rigorous optimization.
of simple structure antennas with enhanced features. Besides,
these antennas can be used in civilian and military platforms. Moreover, the authors have proposed a new palm tree
There are different wireless communication applications structure wideband antenna. The antenna is capable to cover
defined in the electromagnetic spectrum. These applications the 4GHz to 10.4GHz operable frequency range [21]. Kalyan
include WiMAX (3.5GHz-5.8GHz), H-Band (6GHz-8GHz), Mondal et al. demonstrated the inverted question mark
Ultra-Wideband (UWB) (3.1GHz-10.6GHz) and airborne, wideband antenna. The proposed antenna exhibited the good
land and naval radars (8.5GHz-10.5GHz). These frequency gain of 5.5dBi across the frequency span [22]. Further, the
bands are allocated for different wireless communication authors have presented the antennas which were capable to
applications after the approval of Federal Communication cover the different wireless communication applications [23].
Commission (FCC) [3].
Recently, the authors have presented the multiband
Numerous studies about compact broadband antennas for antennas. The different feeding techniques were utilized and
different wireless communication applications have been achieved the multiband characteristics [24] and [25]. A novel
*Corresponding Author.
39 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
miniaturized UWB antenna has been presented. The proposed engraved on the upper side of PGP are used to realize the
antenna has the larger physical dimensions [26]. A high gain improved impedance BW.
tapered slot antenna array was reported in [27]. The authors
have achieved the high gain and substantial impedance BW Moreover, the prototype-II of proposed antenna is depicted
with Wilkinson power divider approach. in Fig. 1(b). It is operated between 3.5GHz-10.5GHz centered
at 6.7GHz performing the operating BW of 7GHz. It can be
In this paper we have designed the two models of the observed that PGP, substrate thickness and the dimensions of
compact antennas and analyze their performance. The antenna L-shape slots have resulted in improved impedance BW. The
design topology is very simple. The proposed antenna proposed antenna is designed and simulated by the EM solver
dimensions are calculated with the standard formulation. HFSS version 13.0.
Moreover, the parametric study of different variables has been
carried out. The performance of antenna parameters at The approximated initial calculated values have been
multiple resonances is observed. The simulation results of obtained from the equations explained in this section. After
prototype-I (reference antenna) and prototype-II (proposed multiple experimental simulations the optimized values of the
antenna) is compared. The broadband fractional impedance designed antenna are listed in Table I.
BW of 100% at 10dB return loss has been observed. The values of dimensions of patch is calculated by the
Moreover, substantial impedance BW improvement of 10.7% following equations (1) and (2), respectively [28].
is analyzed. Finally, the proposed antenna design achieves
near monopole like stable radiation pattern, maximum gain F
and strong current distribution across the surface of the Lp 1
2
antennas. 2hs F 2
1 ln 1.7726
The key contributions of this manuscript are explained as:
r F 2hs (1)
The designed models of the antennas exhibit the
broadband impedance BW, good gain, strong current Lp Lp
distribution and stable radiation pattern across the
standard planes.
Wp
Wp
The proposed antennas possess the compact physical
and electrical dimensions.
Wf
Ls
The organization of the paper is categorized mainly in five
sections. The antenna layout and mathematical strategy covers
in Section II. The optimization of the variables and analysis of
results such as peak realized gain, return loss, radiation pattern
Lf
The reference and proposed antenna model and their (a) (b)
visualization from the front, back and side view perspective Feedline
are depicted in the Fig. 1(a)-(c). The antenna designs are Radiating Patch
composed of compact patch; microstrip feed line and partial
ground plane (PGP). A simple shape of radiator is engraved hs Dielectric Material FR4-Epoxy (Ɛr)
on the top surface of thick substrate with compact dimensions
31.7×19×1.4 mm3. Low cost FR4 Epoxy laminate is used as
substrate material with dielectric relative permittivity value Partial Ground Plane
εr=4.4 and dielectric loss tangent δ=0.02. Moreover, the
proposed antenna is feed by 50Ω simple microstrip feeding (c)
line. The antenna is composed of three layer sheets, i.e. the Fig. 1. (a) Top View of Prototype-I (b) Bottom View of Prototype-II (c)
first layer sheet is dielectric substrate, the second layer sheet Lateral View of the Proposed Antenna.
consists of the compact patch and feeding line etched on top
of the laminate and the third layer sheet covers the PGP etched TABLE. I. PROPOSED ANTENNA DEFINED VARIABLES
on the back side of the laminate. These all elements are made Variable name Values (mm) Variable name Values (mm)
up of copper clad material. The feed line has the dimension of Lp 9.5 LPGP 14.25
16.6×2 mm2 which have great influence to achieve the proper Wp 9.5 WPGP 19
impedance matching. The PGP is taken as optimized value for
Lfl 16.6 Ls 31.7
broader impedance BW. The variables of the radiator are
Wfl 2.0 Ws 19
adopted to adjust the return loss. Moreover, L-shape slots
40 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
8.79 10 9
F
fr r -10
(2)
|S11| (dB)
In the above equations (1) and (2) variable hs represents
the thickness of dielectric substrate εr is relative permittivity,
F represents the wavelength of substrate and fr is resonant -20
frequency.
Moreover, feedline width can be calculated by using the
standard numerical equations (3) and (4), respectively [29]. -30
A
W fl 8e Lf=16mm Lf=16.6mm
hs e 2
2A
(3)
2 4 6 8
Frequency (GHz)
10 12
Where variable A can be calculated as: Fig. 2. Impedance Matching Analysis Related to Feed Line Length.
1
Z 1 2 1 0.11
0
A 0 r r 0.23
Wp=9.1mm Wp=9.2mm Wp=9.3mm
60 2 r 1 r Wp=9.4mm Wp=9.5mm
(4)
-10
Z0 is the characteristic impedance.
|S11| (dB)
III. SIMULATED RESULTS AND ANALYSIS
In this section impedance matching performance related to -20
variables used in the antenna design is explained. Moreover,
the results of return loss (S11), peak realized gain (dBi), surface
current distribution (Jsurf) and radiation pattern are also
discussed and analyzed. -30
A. Parametric Study
This section investigates the impact of the feeding line 2 4 6 8 10 12
Frequency (GHz)
length (Lf) and width of patch (WP), Effect of PGP length
(LPGP) and width (WPGP). These effects realize the matching Fig. 3. Impedance Matching Analysis Related to width of Patch.
performance of proposed antenna. Moreover, parametric study
of the proposed antenna in terms of variables is accomplished 2) Variation of PGP with respect to length (LPGP) and
by running the multiple times rigorous simulations. The width (WPGP): Length of the PGP plays a vital role to achieve
effects of different values of variables are observed. Finally, broadband impedance BW. The PGP dimensions are almost
the optimized values are chosen to validate the proposed half of the dimensions of the dielectric substrate. Fig. 4(a)
antenna prototype. demonstrates the optimum matched result of return loss
1) Variation in feedline(Lf) and Patch (Wp): Microstrip S11<10dB at 14.25mm.
feeding line is key part of the proposed antenna. The antenna Fig. 4(b) shows the optimized value of the width of PGP at
radiator can be excited with the feeding line. It is very 19mm. It is observed that the change in the dimensions of
important to set the proper dimensions of feed line in order to PGP results in the wide impedance BW and proper impedance
achieve the perfect impedance matching. Fig. 2 shows the matching. Finally, the simulation results of radiating patch,
different optimetric values of feedline length ranges from feed line and PGP shows the optimized antenna design
16.2mm to 16.6mm. It is analyzed that the proposed antenna geometry covers the different wireless communication
achieves the good matching performance at 16.6mm value. application.
Moreover, the dimensions of patch also influences over the B. Return Loss (S11)
impedance matching of the proposed antenna. Fig. 3 Fig. 5 delineates the return loss of prototype-I (reference
demonstrates the variation of patch width from 9.1mm- antenna) and prototype-II (proposed antenna) across the
9.5mm. It is analyzed that the optimized values for length and operable frequency range. It is analyzed that the reference
width of patch is achieved at 9.5mm. antenna achieves the broadband impedance BW of 6GHz at
41 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
|S11| (dB)
10dB return loss. Furthermore, it is analyzed that proposed
antenna has the three resonances at 3.9GHz, 6.6GHz and -20
9.8GHz, respectively.
From the above analyzed results of the return loss, it is
concluded that proposed antenna achieved almost 10.7%
-30
improvement in impedance BW as compared to reference
antenna prototype.
C. Peak Realized Gain (dBi)
-40
Fig. 6 shows the peak realized gain of reference antenna 2 4 6 8 10 12
and the proposed antenna.It can be observed at 11.6GHz Frequency (GHz)
frequency reference antenna exhibits peak realized gain of Fig. 5. S11 vs. Frequency Plot of Proposed and Reference Antenna Design.
5dBi. Besides, the multiple resonances such as: at 4.2GHz the
antenna exhibits the gain of 4.1dBi, at 6.5GHz the acceptable Moreover, it is analyzed that the proposed antenna has
gain of 3.45dBi and at 8GHz the gain of 3.6dBi is achieved. achieved the high gain of 5.2dBi at 10.9GHz. However, the
0
multiple resonances such as: at 3.9GHz, the acceptable gain of
LPGP=13.25mm LPGP=13.5mm LPGP=13.75mm 3.7dBi, at 6.6GHz, the gain of 3.2dBi and at 9.8GHz the good
gain of 4.9dBi is observed. Besides, it can also be seen that
from 7.2GHz to 7.6GHz the gain is degraded upto 0.18dBi.
-10 The degradation in the gain is observed because of the L-
shaped slots etched behind the feedline.
|S11| (dB)
42 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
6 0 0
Proposed Antenna Reference Antenna 330 30 330 30
300 60 300 60
Peak Realized Gain(dBi)
4
270 90 270 90
300 60 300 60
43 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
V. FUTURE WORK [13] A. Valizade, J. Nourinia, B. Mohammadi, and P. Rezaei, “New design
of compact dual band-notch ultra-wideband bandpass filter based on
The work presented in this paper can be further extended coupled wave canceller inverted T-shaped stubs,” IET Microwaves,
to create the reconfigurable notch band functions at particular Antennas Propag., vol. 9, no. 1, pp. 64–72, 2014.
frequency band of spectrum. Moreover, a broadband antenna [14] J. Li, Q. Chu, Z. Li, and X. Xia, “Compact Dual Band-Notched UWB
array topology will also be focused by implementing the MIMO Antenna With High Isolation,” EEE Trans. Antennas Propag.,
vol. 61, no. 9, pp. 4759–4766, 2013.
efficient Wilkinson power divider and their performance will
be analyzed in real time environment. [15] A. Mousazadeh, M. Naser-Moghaddasi, F. Geran, S. Mohammadi, and P.
Zibadoost, “Broadband CPW-Fed circularly polarized square slot
antenna with arc-shaped and inverted-L grounded strips,” Appl. Comput.
ACKNOWLEDGMENT Electromagn. Soc. J., vol. 28, no. 4, pp. 314–320, 2013.
Authors Zaheer Ahmed Dayo, Qunsheng Cao, Yi Wang [16] A. Katuru and S. Alapati, “Design and analysis of modified circular
and Saeed Ur Rehman gratefully acknowledge the support patch antenna with DGS for UWB applications,” in Lecture Notes in
from National Natural Science Foundation of China under Electrical Engineering, 2018, vol. 434, pp. 537–545.
grant No. 61871219. [17] A. Kamalaveni and M. Ganesh Madhan, “Halve dumbbell shaped DGS
tapered ring antenna for dual-band notch characteristics,”
REFERENCES Electromagnetics, vol. 38, no. 3, pp. 189–199, 2018.
[1] J. R. Verbiest and G. A. E. Vandenbosch, “A novel small-size printed [18] K. F. Jacob, M. N. Suma, R. K. Raj, M. Joseph, and P. Mohanan,
tapered monopole antenna for UWB WBAN,” IEEE Antennas Wirel. “Planar branched monopole antenna for UWB applications,” Microw.
Propag. Lett., vol. 5, no. 1, pp. 377–379, 2006. Opt. Technol. Lett., vol. 49, no. 1, pp. 45–47, 2007.
[2] A. Dastranj and H. Abiri, “Bandwidth enhancement of printed E-shaped [19] Y. Z. Cai, H. C. Yang, and L. Y. Cai, “Wideband monopole antenna
slot antennas fed by CPW and microstrip line,” IEEE Trans. Antennas with three band-notched characteristics,” IEEE Antennas Wirel. Propag.
Propag., vol. 58, no. 4, pp. 1402–1407, 2010. Lett., vol. 13, pp. 607–610, 2014.
[3] F. C. Commission, “Revision of Part 15 of the Commission’s Rules [20] S. Koziel and A. Bekasiewicz, Multi-Objective Design of Antennas
Regarding Ultra-Wideband Transmission Systems,” First Rep. Order …, Using Surrogate Models. 2016.
no. FCC02-48, pp. 1–118, 2002. [21] S. K. Palaniswamy, K. Malathi, and A. K. Shrivastav, “Palm tree
[4] Y. Sung, “Triple band-notched UWB planar monopole antenna using a structured wide band monopole antenna,” Int. J. Microw. Wirel.
modified H-shaped resonator,” IEEE Trans. Antennas Propag., vol. 61, Technol., vol. 8, no. 7, pp. 1077–1084, 2016.
no. 2, pp. 953–957, 2013. [22] K. Mondal, A. Shaw, and P. P. Sarkar, “Inverted question mark
[5] M. Gupta and V. Mathur, “A new printed fractal right angled isosceles broadband high gain microstrip patch antenna for ISM band 5.8
triangular monopole antenna for ultra-wideband applications,” Egypt. GHz/WLAN/WIFI/X-band applications,” Microw. Opt. Technol. Lett.,
Informatics J., vol. 18, no. 1, pp. 39–43, 2017. vol. 59, no. 4, pp. 866–869, 2017.
[6] M. Ojaroudi, N. Ojaroudi, and N. Ghadimi, “Dual band-notched small [23] M. L. Meena, M. Kumar, G. Parmar, and R. S. Meena, “Design analysis
monopole antenna with novel W-shaped conductor backed-plane and and modeling of directional UWB antenna with elliptical slotted ground
novel T-shaped slot for UWB applications,” IET Microwaves, Antennas structure for applications in C- & X-bands,” Prog. Electromagn. Res. C,
Propag., vol. 7, no. 1, pp. 8–14, 2013. vol. 63, no. April, pp. 193–207, 2016.
[7] M. Naser-Moghadasi, R. A. Sadeghzadeh, T. Sedghi, T. Aribi, and B. S. [24] P. V. Naidu and A. Malhotra, “A small ACS-fed tri-band antenna
Virdee, “UWB CPW-fed fractal patch antenna with band-notched employing C and L shaped radiating branches for
function employing folded T-shaped element,” IEEE Antennas Wirel. LTE/WLAN/WiMAX/ITU wireless communication applications,”
Propag. Lett., vol. 12, pp. 504–507, 2013. Analog Integr. Circuits Signal Process., vol. 85, no. 3, pp. 489–496,
[8] M. Koohestani and M. Golpour, “U-shaped microstrip patch antenna 2015.
with novel parasitic tuning stubs for ultra wideband applications,” IET [25] Z. A. Dayo, Q. Cao, P. Soothar, M. M. Lodro, and Y. Li, “A compact
Microwaves, Antennas Propag., vol. 4, no. 7, p. 938, 2010. coplanar waveguide feed bow-tie slot antenna for WIMAX, C and X
[9] A. T. Mobashsher, M. T. Islam, and N. Misran, “Wideband compact band applications,” in 2019 IEEE International Conference on
antenna with partially radiating coplanar ground plane,” Appl. Comput. Computational Electromagnetics (ICCEM), 2019, vol. 26, no. 3, pp. 1–3.
Electromagn. Soc. Newsl., vol. 26, no. 1, pp. 73–81, 2011. [26] L. Guo, M. Min, W. Che, and W. Yang, “A Novel Miniaturized Planar
[10] M. C. Tang, T. Shi, and R. W. Ziolkowski, “Planar ultrawideband Ultra-Wideband Antenna,” IEEE Access, vol. 7, pp. 2769–2773, 2019.
antennas with improved realized gain performance,” IEEE Trans. [27] P. Soothar, H. Wang, B. Muneer, Z. A. Dayo, and B. S. Chowdhry, “A
Antennas Propag., vol. 64, no. 1, pp. 61–69, 2016. Broadband High Gain Tapered Slot Antenna for Underwater
[11] D. Zhang and Y. Rahmat-Samii, “Top-cross-loop improving the Communication in Microwave Band,” Wirel. Pers. Commun., no.
performance of the UWB planar monopole antennas,” Microw. Opt. 123456789, 2019.
Technol. Lett., vol. 59, no. 10, pp. 2432–2440, 2017. [28] C. A. Balanis, Antennas Third Edition, vol. 45, no. 3. 2005.
[12] Y. Sung, “Bandwidth enhancement of a microstrip line-fed printed [29] D. M Pozar, Microwave Engineering, 3rd Edition. 2004.
wide-slot antenna with a parasitic center patch,” IEEE Trans. Antennas
Propag., vol. 60, no. 4, pp. 1712–1716, 2012.
44 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In software development projects, the process of to express their feedback and experiences online via social-
requirements engineering (RE) is one in which requirements are network sites, forums, and blogs.
elicited, analyzed, documented, and managed. Requirements are
traditionally collected using manual approaches, including Because of changing contexts and user needs, continuous
interviews, surveys, and workshops. Employing traditional RE requirements elicitation should be adopted to ensure that
methods to engage a large base of users has always been a requirements stay refreshed and that needs are addressed [7].
challenge, especially when the process involves users beyond the Stakeholder needs and technologies change over time,
organization’s reach. Furthermore, emerging software exacerbated by the rise of crowd-generated data. Automated
paradigms, such as mobile computing, social networks, and cloud requirements elicitation methods and analysis should be
computing, require better automated or semi-automated incorporated to enable requirements engineers to acquire and
approaches for requirements elicitation because of the growth in analyze online data efficiently. Automation facilitates access to
systems users, the accessibility to crowd-generated data, and the online crowd-generated data and the use of these data for
rapid change of users’ requirements. This research proposes a systems’ improvements [7]. Automated or semi-automated
methodology to capture and analyze crowd-generated data (e.g., requirements elicitation approaches should be able to overcome
user feedback and comments) to find potential requirements for issues facing existing traditional approaches [8]. This research
a software system in use. It semi-automates some requirements- proposes a methodology that collects crowd-generated data
elicitation tasks using data retrieval and natural language
from social networks (e.g., Twitter) and processes the data
processing (NLP) techniques to extract potential requirements. It
using natural language processing (NLP) techniques to extract
supports requirements engineers’ efforts to gather potential
requirements from crowd-generated data on social networks
potential emerging requirements for a certain software product.
(e.g., Twitter). It is an assistive approach that taps into unused The rest of the paper is organized as follows. Section 2
knowledge and experiences emphasizing continuous presents the literature review while Section 3 details the
requirements elicitation during systems use. proposed methodology and its supporting tool. Sections 4 and
5 respectively present the discussion and the conclusion of this
Keywords—Requirements engineering; RE; crowd data
study.
mining; NLP; Twitter; continuous requirements elicitation
II. LITERATURE REVIEW
I. INTRODUCTION
The success of a system development or an upgrade
Requirements engineering (RE) is the process of collecting,
depends on a well-developed RE process that successfully
defining, documenting, and maintaining the requirements of a
elicits and manages stakeholder requirements, resulting in a
software system [1]. It is fundamental during the software
higher level of satisfaction [4, 9]. Requirements elicitation is
development cycle to obtain users’ needs by utilizing effective
traditionally the first phase of obtaining requirements.
means of requirements elicitation, analysis, and management
Elicitation is the most important phase because the collection
[2]. Getting the requirements right is important because
of poor requirements can lead to project failure [10-12]. The
mistakes cascade to subsequent development stages. Owing to
involvement of users and customers in the RE process leads to
poor RE practices, deficiencies at this phase cost more later
many benefits, such as improved system acceptance, more
and often result in systems failure [2-4].
accurate and complete requirements, and improved project
Traditionally, elicitation is done at the beginning of success rates [13]. Many issues lead to poorly collected
software development. Recent approaches have advocated requirements, such as ambiguous project scopes, poor system
continuous requirement elicitation to capture user feedback and understanding, and volatility where the evolved users’ needs do
experiences during system’s use [5]. Elicitation is needed not meet the original requirements [14].
during system’s use to understand new feature requests, issues,
Various requirements elicitation approaches have been
and emerging requirements [6]. Requirements elicitation for
suggested [15]. Most existing techniques are manual and
traditional software systems has been well-studied, but new
assume the presence of the stakeholders involved. Employing
computing paradigms (e.g., social media, mobile apps, and
such techniques can rapidly become expensive and resource-
cloud computing) require different assumptions and
intensive, particularly when dealing with larger stakeholder
approaches [5]. These new computing paradigms enable users
populations [16-18]. Employing such techniques to engage a
large user base has always been a challenge, especially when
45 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
there are large numbers of software users beyond the predefined patterns from user manuals and project reports. The
organization’s reach [7]. Traditional RE approaches ignore text in these documents was tokenized and POS tagging was
opportunities to continuously engage large and heterogeneous used to annotate the text. Then, topic modeling was applied to
groups of users who express their feedback on social networks group-requirement items of similar content to avoid
and other websites. Better approaches are needed to tap crowd- information overload. Whereas it is considered appropriate to
generated data (e.g., feedback and opinions) to enable reduce the burden of gathering requirements from scratch,
developers to consider them when developing their product’s some limitations have been reported, such as unclear extracted
next version [7]. requirements and lack of comprehensive patterns [25].
Stakeholder goals, environments, technology evolvement, Furthermore, an approach was developed to automate some
and the emergence of new computing paradigms require requirements elicitation tasks using a tool that gathered
continuous requirements elicitation. For example, social- stakeholder input in a centralized repository. Then, it used
network sites and mobile applications generate data that can be extended markup language and extensible stylesheet language
collected and analyzed for potential requirements [7]. The rise transformations to render specifications [26]. In [27], a method
of social networks and mobile applications has enabled the was suggested to extract requirements from textual data in
collection of massively generated crowd data. Social-network documents. NLP techniques (e.g., tokenization, POS tagging,
users can contribute their feedback directly or indirectly and clustering) were used. Another study examined similar
regarding system improvements [19, 20]. Whereas social project documentation to extract potential requirements using
networks were not designed for the purpose of requirements NLP techniques (e.g., POS tagging) [28]. Another approach
engineering, many companies include social networks in their was proposed that used online customer reviews to extract
software development process for this purpose [21]. needs and preferences regarding a specific product [29].
Understanding public opinion and demands is a time- III. PROPOSED CONTINUOUS REQUIREMENTS-ELICITATION
consuming process because of the high volume of crowd- METHODOLOGY
generated data that must be reviewed [22]. Thus, automatic
approaches to elicit and analyze such data are needed to This section outlines the proposed methodology and its
achieve faster response times [7]. Automation facilitates the supporting tool. This research provides an approach to
identification and analysis of potential requirements that are automatically collect crowd-generated data via Twitter and
otherwise challenging and unreachable using traditional RE process it using NLP techniques to find requirements. The
[8]. proposed methodology is shown in Fig. 1. It enables engineers
to elicit data from Twitter and analyze it using NLP techniques
An emerging theme within RE research is Crowd-based to find potential requirements. Twitter was selected because it
requirements engineering (CrowdRE). It is an overarching term is a popular microblogging social-media network and a
for the employment of automated or semi-automated methods potential data source to extract requirements [30, 31].
to elicit and explore data from a crowd to derive potential
requirements [7]. Crowdsourcing in requirements elicitation The methodology has four main steps: tweets collection
would enable the continuous requirements elicitation process and filtering, applying POS tagging, requirements generation,
during the life cycle of the software product. Such a practice and requirements clustering. The following subsections
would facilitate a deeper, wider, and more up-to-date illustrate the proposed methodology steps and its instantiation
perspective of how users perceive systems and to understand using AutoReq.
how requirements evolve [5]. Typically, a crowd is a large and A. Tweet Collection (Pattern Matching) and Filtering
heterogeneous group of existing or prospective users [7].
AutoReq enables requirements engineers to input a search
CrowdRE captures and analyzes user needs regarding the
keyword (e.g., the name of an existing system) and search
evolution of existing software systems, and it monitors
twitter feed. The tool uses the Twitter application program
software system usage and experiences. Crowd users report on
interface (API) to retrieve real-time tweets matching the search
a variety of aspects, such as problems, improvements, or
criteria (i.e., predefined pattern). For example, if we were
extension ideas, which are useful for software development
interested in finding the feedback of an existing system, X, the
teams [7].
patterns added to AutoReq would include ―X should …,‖ ―X
Crowd-generated textual data should be retrieved and could …,‖ and ―X lacks …‖.
processed with NLP techniques. NLP concerns the application
In this study, the software system of interest is Snapchat. It
of computational techniques for automatic parsing, analysis,
is a global multimedia messaging application. It was selected
and representation of human language. Many techniques have
because it is widely used and has very diverse user groups with
been suggested to process raw text in natural languages. For
constantly evolving requirements. Prior to this experiment, we
example, tokenization is a technique used for splitting a stream
noticed users tweeting potential requirements, additional
of text into its basic elements (i.e., tokens) such as words and
features, complaints, and other issues about Snapchat. An
phrases and other symbols [23]. Part-of-speech (POS) tagging
AutoReq pattern search list was used. Then, tweets were
is used to assign labels (e.g., noun, verb) to each identified
filtered from unwanted noise (e.g., hashtags, user mentions,
token in a given text [24].
and universal resource locators). They were then saved to the
Several studies have attempted to automate the AutoReq database. During the active stream retrieval of tweets,
requirements elicitation process using NLP techniques. For more than 350 tweets having the word ―Snapchat‖ were
example, NLP was used to extract early requirements matching retrieved, and only 47 matched the predefined pattern.
46 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
47 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Part-of-Speech (POS) Tagging developed based on the qualitative analysis and the use of
Retrieved tweets were then tokenized by breaking them existing conjunction words in English [32]. Using recursion,
into tokens. Each tweet was then annotated using the Stanford tweets were split into parts using a conjunction word. Each part
POS tagger [24]. Tags were assigned to each word, depending of the tweet was processed alone, and then both parts were
on its role in the sentence. Still, there was some incorrect combined as one requirement using the format ―Snapchat shall
tagging. For example, the word ―update‖ was incorrectly + combined tweet output.‖
tagged in the tweet ―snapchat should remove the last update.‖ D. Requirements Clustering
It should have been tagged as a noun, but it was instead tagged
as a verb. This phenomenon can lead to the generation of After requirements generation, clustering can be useful,
confusing requirements. particularly in cases where the retrieved tweets are large.
Generated requirements were clustered to provide an
C. Requirements Generation aggregated perspective of common themes from the generated
After tagging the words of each tweet, the first annotated requirements. Clustering was conducted using the RxNLP
verb and the closest three words were used to generate a sentence-clustering API [33]. It groups text tokens on a
requirement clause. Using a predefined requirement template sentence level. It can be applied to short texts, or, in this
within the tool, the requirement phrase was structured as ―X research, tweets, to build logical and meaningful clusters with
shall + requirement clause.‖ In this experiment, generated suggested topics for each cluster.
requirements were structured as ―Snapchat shall + requirement Generated requirements were clustered based on the most
clause.‖ In some cases, a tweet contained more than one frequent topic themes, making it easier to find requirements of
sentence. Thus, a recursion function of the tool was used to interest. The results, as shown in Fig. 2, contain the cluster
process the second part of the tweet. To find common topic, cluster score, and cluster tweets. The cluster topic is a
conjunctions that potentially indicate the need for the use of the suggested name of the cluster contents, whereas the cluster
recursion capability, a qualitative analysis of the raw collected score describes the topic meaningfulness and cluster size. It
tweets was conducted. Then, the connection-words list was facilitates cluster ranking and unwanted cluster pruning.
48 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
IV. DISCUSSION techniques to detect the most frequent words in their data
source to build requirements [25].Some studies used clustering
RE mostly uses traditional data sources (e.g., forms, to cluster the requirements based on predefined centroids,
reports, notes, workshops, and meetings) and manual regardless of similarity [27].
approaches such as interviews for capturing stakeholder
requirements. The wide use of social networks and mobile apps V. CONCLUSION
has contributed to a massive growth in online crowd-generated Requirements elicitation is a crucial phase in the software
data. Crowd users report on a variety of issues based on development life cycle designed to fully understand users’
software problems, desired improvements, and extension ideas, needs. During the elicitation process, interviews, workshops,
which are potentially useful for software development teams reports, and manuals are typically used to generate
[7]. These data are often massive, unstructured, and manually requirements. However, emerging computing paradigms and
inaccessible [22]. Hence, recent research has called for the the massive growth of crowd-generated data require automated
development of automated approaches to capture and analyze elicitation approaches. Crowds directly or indirectly express
these data to locate potential requirements. A rising their feedback, comments, and opinions regarding an existing
opportunity for RE lies within the use of hidden and unused system on social networks and similar platforms. Gathering
crowd-generated data [7]. data using existing requirements elicitation techniques is an
This research endeavored to explore this research area and arduous process, particularly when dealing with large-scale
contributed as follows. First, this study is early research systems.
exploring the use of crowd-generated social-networks data to This research proposed a methodology and proof-of-
find new requirements for an existing software system. It concept to automate the retrieval and analysis of crowd-
proposed a methodology and a tool to capture and analyze generated data from Twitter using NLP techniques to find
crowd-generated data to identify potential requirements. Such potential requirements of an existing software product. This is
an approach is needed to achieve fast responses to user needs an early study investigating the use of crowd-generated data to
and to explore the hidden, unused data generated by users on find potential requirements. It employs NLP techniques to
social networks [7]. The developed methodology and tool automatically analyze captured textual data, and it enables a
support requirements engineers in their tasks of monitoring and continuous requirements elicitation process during the use of
eliciting potential requirements from crowd-generated data software products. It also uses a clustering-sentence technique
using their reported feedback, comments, and experiences. to cluster requirements based on their similarity to automate
Second, this research used NLP techniques to automatically grouping of similar tweets. This reduces manual RE efforts.
analyze the captured textual crowd-generated data (i.e., Because every research effort is limited, there are some
tweets). NLP techniques support requirements engineers by limitations with this study. First, because we proposed a semi-
automating parsing, analysis, and representation of textual data. automated tool, there needs to be an RE verification and
Manual inspection, filtering, and processing are time- evaluation of the generated requirements to assess their
consuming and resource-intensive. Thus, an automated relevance and importance. Second, this study inherited some
approach of crowd-generated data retrieval and processing limitations of the applied NLP techniques, particularly POS
reduces time and resource utilization. Nonetheless, there were tagging. In addition, automated text processing and analysis
some issues with unclear generated requirements phrases from have their own limitations. For example, there were some
incorrect tagging when using a POS tagger. To overcome this, generated requirements that were not meaningful because of
the developed tool was designed to show the original tweets either incorrect tagging or retrieval. Another limitation was
and the generated associated requirement phrases to help inherited from the data source, owing restricted access to
requirements engineers trace and understand the generated tweets using the Twitter API and the 140-character limitations
requirements. at the time of the execution of the experiment.
Third, this research emphasized the continuous In the future, extra efforts are needed to improve the
requirements elicitation process over a software product life suggested approach. For example, additional NLP techniques
cycle using crowd-generated data [5]. Feedback and (e.g., collaborative filtering) should be included to extract
experiences of current or prospective users were continuously relevant requirements. Furthermore, richer data sources are
captured about new features, emerging needs, and other issues. suggested, including Facebook and online app reviews, to
This approach is not easily implementable with traditional data collect richer requirements. In general, further research is
sources, such as manuals and reports. needed to develop methods and tools that facilitate continuous
Fourth, a sentence-clustering technique was used to cluster requirements elicitation to retrieve and analyze online crowd-
requirements based on their similarity [33]. In this research, generated data during software systems use.
every processed tweet was treated as a unique requirement and
a genuine idea that may lead to redundant requirements. Thus, ACKNOWLEDGMENT
a sentence-clustering technique was used to enable The authors extend their appreciation to the Deanship of
requirements engineers to look at clusters when the generated Scientific Research at King Saud University for funding this
requirements are large. This reduces requirements engineers’ research.
manual efforts. Previous research mostly used topic modeling
49 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
50 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Now a days, there is demanding needs of Traditional methods of study are more complex, boring
developing interactive mediums of study. As our conventional and time taking. The field Human computer interaction directs
methods of learning are not very effective. Programming has us to how to make an interactive application? What are the
become one of the core subjects of every field of study due to requirements of user? What are the problems faced by users?
their vast use. However, introducing computer programming to What the user needed? There are some questions invoke at
those students who’s are not familiar with programming is a every developer while he/she build an application of software.
tough task. Use interactive learning through visual effects using HCI gives the better ways to develop an interactive
AR (Augmented Reality) developed to provide a platform for application. Students’ who learn in an e-learning environment
new students to interact more in learning environment. As this
would be independent of distance, time, computing platform
learning environment becomes more effective it is easier for new
comers to understand key concepts of programming more
as well as classroom size also.
effective way. The knowledge needed to write a computer program is an
important part of literacy in modern society. While private
Keywords—Augmented reality; object-oriented programming; coding classes are expensive and limited.
unity; visualization; human computer interaction; Vuforia;
rendering; compiler In programming some of the most difficult things to
understand are concepts that we use to develop data structures
I. INTRODUCTION and algorithms. Sometimes even for the more intelligent
Application will allow the people to visualize the OOP people they cannot understand some concepts due to the fact
concepts and check their interactions and effects to understand that each person has a different mind-set.
them easily with efficiency rather than just sitting and thinking This application will allow such people to visualize such
about them by using and visually seeing them their rate of concepts and check their interactions and affects to understand
progress will increase. them easily with efficiency rather than just sitting and thinking
Although there is lot of e-learning application available on about them by using and visually seeing them their rate of
online platforms but there is no special application for object- progress will increase.
oriented programming. This project consists of a web Although there is lots of e-learning application available
application and an android application. Web portion is on online platforms but there is no special application for
consisting of textual based tutorials for learning and android object-oriented programming. This project consists of a web
application is for visual learning. application and an android application. Web portion is
II. LITERATURE REVIEW consisting of textual based tutorials for learning and android
application is for visual learning.
The importance of learning through computer starts from
21th century and their importance gradually increase with the Originally this approach of learning started in the more
passage of time. Programming is one of basic course required developed countries of the world such as U.S.A and U.K in
in any computer science related field. For most of the students, these countries to improve the education system e-learning
it is also one of the most challenging tasks that how to was introduced now though with the advent of augmented and
understand coding or how to write a computer program. For virtual reality, this system of e-learning is being taken a step
improving the novice experience with learning to program, further by making learning more interactive and hence more
visual environments and effects can be designed. At university interesting by using such technologies. In such developed
level as well as the industry have placed increasing countries such application has started being widely used in
importance on the early experience of students to object- young children [3], hence our aim is the same to increase
oriented programming (OOP). education efficiency by using these technologies.
There is a need to develop an interactive e-learning Moreover, The School Education Department of Punjab
environment for students to overcome the challenge of learn has been working closely with Punjab Information
programming. Incredible developments had been occurring in Technology Board (PITB) to develop and implement E-
computer technology and its availability in that time duration. learning solutions for secondary and higher secondary
Computer technology used for learning in schools, students in which, PCTB textbooks for Grades 6 through 10
universities, business market, industries and the many other have been digitised and augmented with thousands of
professions in last ten years. So, the number of learning interactive multimedia components like Animations,
software’s and applications has growth dramatically. Simulations, 3D models and videos. [2]
51 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
More related works would be those that follow the four code also available through compiler. User with be able to
steps [4], to teach students through e-learning, due to the fact discuss or ask any question about the topic in comment
that much of the increase in understanding comes from use of section. Because it’s an interactive web application so tutorials
what is known as "visual literacy." Visual literacy is defined are locked once someone take a tutorial about some topic the
as the ability to understand data and information presented in a next task is to give a simple multiple-choice quiz. After
pictorial format or graphic images [1], Visual literacy has been successfully completed the quiz then the user moves towards
proven to enhance learning especially in subjects where they next tutorial and so on.
do not have much prior knowledge. [5].
B. Compiler
In fact, studies show that using graphics in instructional Compiler is basically for try yourself section that user pic
modules promotes greater understanding in general. It is in example code of some specific topic and makes changes in it
fact proven that less is more beneficial if it can be better as shown in Fig. 2: Compiler able to work with multiple
understood [8], Logic building process is also a tough task for classes of object-oriented programming. As in case of any
students and many students confused about different wrong code written by user Errors also shows with line
approaches to solve some programming problem. These number on console panel.
aspects would be challenging for teachers and students [6],
and the learning of the application of what is being learnt C. Quiz
while it is being learnt is important [7], Researcher shows the Fig. 3 shows quiz section consists of multiple choice-based
object-oriented concept. Some concepts students understand questions. Answers will be provided also for confirmation of
properly but some concept cannot understand. It finally points selected choice. A progress bar will be saving the quiz data of
difficulties in some areas and result, planning to give an easy users.
and effective way to teaches OOP [9], by participating in
applied task, the comprehension of the subject becomes better
[10].
It is known through research that the programming skills
of first year students are not up to the required level as most of
them have not been exposed to how to solve tasks and do
programming [11] hence in the past there have been many
application that have focused on teaching beginners how to
code but in most of these applications the user only learns the
syntax and does not have an editor to practise it side by side
and while it is important to know how to build a solution [12]
it is also better to at least once make the solution as well Fig. 1. Explainer
because it cements our understanding, there have also been
many lone editors for the java language but they are
complicated and not very good for beginners to work on,
while in the system we are proposing we will try to combine
both an explainer as well as an editor with visual
representations to allow the user to quickly understand
concepts and learn the language efficiently as in interactive
setting the user unknowingly tries harder [13].
III. INTERFACES
Augmented Reality followed basic human interaction
usability principles in user interface creation. Consistent Fig. 2. Compiler.
interface and design present across all modules. By default,
augmented reality opens in full screen mode in landscape
orientation on some specific devices those who have sensors
to show augmented objects. Interface is compatible with cross
devices. Based on human computer interaction methods, we
design interfaces very user friendly.
Following are prime interfaces components.
A. Explainer
Explainer is mean interface component of project. This
module consists of basic concepts tutorials of object-oriented
programming. Every tutorial consists of three sections such as
topic name, topic explanation and try yourself code as shown
in Fig. 1. We provided the code examples of every topic
which is easy to understand and make changes in practice Fig. 3. Quiz.
52 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
D. Rendering Screen
Majority part of user interface cover by Rendering screen.
Rendering screen helps us to seen Augmented reality (AR)
components. One important thing that AR not supports with
all the mobile devices, but some specific mobile devices
shows AR object through camera to see elements present in
surrounding.
IV. METHODOLOGY
Augmented Reality follows visual based interactive
learning method to teach programming. The field of Human-
Computer Interaction (HCI) or in other words show something
by using visual effects is focused on enhancement the
effectiveness as well as efficiency of human-computer
interfaces through the development of both hardware and
software designs to recognize human characteristics and
behavior. Advancements in HCI technology can lead to
enhanced Augmented reality (AR) experiences by providing
more natural like environment and efficient methods for a user to
interact with a real or virtual environment in an effective way. Fig. 5. Compiler.
A. Software Architecture
Fig. 4 shows software architecture. According to software
architecture user interacts with web application or android
application and moves towards explainer section for learning
and editor section to explore learning concepts and tryout code
examples with the help of compiler. All the learning data
managed by an admin and only admin have authority to add,
update or delete data in database. User might be suggesting
some things in comments section as feedback.
B. Compiler
We make an API in java for makes an online compiler
which code shows in Fig. 5 and integrate it to Laravel Fig. 6. Augmented Reality App (Car Model (1)).
framework. We used library “Tools and Diagnostics” for
multiclass compiler. This compiler saves all the classes into an
array and separated with Java API basically is a JAR file
which is run by using CMD.
C. Augmented Reality (AR) apps with unity and Vuforia
We can use AR to teach the concepts of OOP in more
optimal manner. Augmented Reality can be used to teach OOP
concepts. Inheritance can be taught using AR based mobile
devices. Similarly, the concept of composition, aggregation,
polymorphism etc. also taught by using AR app. We prepare
the models of CAR which shows in Fig. 6, 7 and HUMAN
which shows in Fig. 8. Firstly, we show that how to initialize
Fig. 7. Augmented Reality App (Car Model (2)).
an object in OOP then we move towards other major concepts
of OOP. Make models of AR in unity 3D & Vuforia.
53 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
V. RESULTS AND DISCUSSION allow for more immersion and not let the user get bored. This
According to studies, it is observed that how augmented application can be further extended to the professional level.
reality can be useful in teaching prospective such as for OOP This can also be really helpful for the beginners. More over
concepts. It is more practical way to teach students in an artful the system can with some time include its own compiler that
environment. It is interesting and growing field. But lack of can visually show objects which will form basis for a
resources, awareness and devices that field not the part of simulator for testing running of the application.
studies even in modern institutes. REFERENCES
[1] Suzanne Stokes. Visual Literacy in Teaching and Learning: A Literature
Augment reality gives us optimal solutions. By using Perspective. Electronic Journal for the Integration of Technology in
methods and techniques of AR will make revolutionary Education, vol. 1, no. 1.
changes in traditional learning process and enhance the [2] https://elearn.punjab.gov.pk
modern learning techniques. The main purpose of this [3] Tanvi Patel. Learning Object Oriented Programming Using Augmented
research is to find out the ways in which AR can be used in Reality: A Case Study with Elementary School Students
studies and we focus on OOP concepts and try out to develop [4] Hinterberger, H., 'E-Learnig: Make it as simple as possible, but not
a system which is more user friendly. We initially targeted simpler'. Afr. Technol. Dev. Forum Journal, Vol. 4, Issue 2, July, 2007.
small amount of population because everyone has not AR [5] Chanlin, L. (1997). The effects of verbal elaboration and visual
supported devices only few mobiles with specific sensors elaboration on student learning.
would be able to run an AR based application. [6] International Journal of Instructional Media, 24(4), 333-339. Retrieved
December 26, 2001, from EBSCOhost database (Academic Search
VI. CONCLUSION Elite).
[7] Heinich, R., Molenda, M., Russell, J. D., & Smaldino, S. E. (1999).
Augmented reality is a new concept but can use generate Instructional media and technologies for learning (6th ed.). Upper
optimal results. In the education field, it can be used in a very Saddle River, NJ: Prentice-Hall.
effective manner as it can be used to teach some concepts in a [8] Mayer, R. E., Bove, W., Bryman, A., Mars, R., & Tapangco, L. (1996).
more practical way to students. Use of AR will make When less is more: Meaningful learning from visual and verbal
revolutionary changes in traditional teacher entered teaching summaries of science textbook lessons. Journal of Educational
Psychology, 88(1), 64-73.
process in future.
[9] Kelleher, C. and Pausch, R. Lowering the barriers to programming: A
VII. FUTURE WORK taxonomy of programming environments and languages for novice
programmers. ACM Computing Surveys 37, 2 (June 2005), 83–137.
The project can be extended to theoretically all the [10] Tony J. A Participative Approach to Teaching Programming, ACM
programming languages as this is in essence a tool for e- SIGCSE, Volume 30 Issue 3, Sept. 1998, Pages 125-129 1998.
learning using visual techniques, in the future, this application [11] M. McCracken, V. Almstrum, D. Diaz, M. G. anD. ianne Hagan, Y. B.
can be enhanced to follow “one size fits all” concept. Web Kolikant, C. Laxer, L. Thomas, I. Utting, and T. Wilusz. A multi-
Programming as well as Database Programming can also be national, multi-institutional study of assessment of programming skills
of first-year CS students. SIGCSE Bulletin, 33(4):125– 180, 2001.
included as in databases, we also require large amounts of
[12] Elliot Soloway. Learning to Program = Learning to Construct
visualizations and have to ponder upon the many what ifs. Mechanisms and Explanation,1986.
Machine learning techniques can be used to provide exercises
[13] Malone, T. (1980). What makes things fun to learn? Heuristics for
to users based on their progress. This application can be made designing instructional computer games. Proceedings of the 3rd ACM
in game type theme in which user can progress only by SIGSMALL Symposium and the 1st SIGPC Symposium (pp. 162–169).
completing a previous task to a satisfactory level. This will Palo Alto, US.
54 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—As people are in the era of the web, most of the II. THE ASPECT OF E-GOVERNMENT STRUCTURE
society is using networks in their daily task, governments had
found, it is crucial to build an electronic entity that was named e- Governments are always searching to save costs in
government, to make transactions easier for citizens, and to providing services chiefly along with the economic crisis
make government nearer to society. The objective of this study is existences. In this line, e-government possesses vital benefits
to assess the extent of e-government application on different to developing nations such as citizen empowerment by
countries, particular in Jordan; in addition, several experiences offering them with a variety of options online channels to
were displayed in this study. The examination was qualitative improve the quality of service delivery [4]. A review of
through interviewing governmental employees for extracting present literature exposes that Jordan has been heavily
results based on their answers, focusing on the continuity of investing in enhancing its e-government facilities to the
using e-government by users as a dependent variable. The community.
conclusion was that the policies are trending to build their e-
government entity, and to make it available for citizens to use. E-government is planned the procedure of cooperation
Further, this study recommends the government to concentrate among government as well as society. According to [5], one
on the path of building individuals’ trust as well as using social critical aspect in the execution of e-government
influence to reinforce the idea of e-government service and evolve administrations is the acknowledgment and the readiness of
its usage. society to utilize e-government administrations. From citizens’
viewpoint, e-government allows persons to access public
Keywords—e-Government; citizens; governmental transactions; services to do their government transactions in a convenient
Jordan way at anytime from anywhere.
I. INTRODUCTION E-government is an innovative form of advanced
The improvement of ICT is quick as of now significantly technology, with a series of set processes that the citizens' logs
affect human life. In the line with this development, the into the website by using a private username and password
worldwide versatile Internet clients' infiltration has achieved based on their selection and secret key with the end goal to do
4.68 billion by 2019 [1]. In Jordan, for instance, the their various online transactions. An outline of the design of
penetration rate has come to 83 percent. This provoked an the e-government is illustrated in Fig. 1, which links to the
adjustment in procedures, capacities, and approaches in essential beneficial components and their jobs inside the
various areas of human life be found on ICT, including open framework. Connection among citizens and specialist system
segment administrations. for government frameworks is upheld by staggered exchanges.
Furthermore, Jordan has a tremendous successfully Furthermore, the e-government framework bolsters
mounting in telecommunication industry in the recent years. correspondence with different servers, for example, Internet
The Central Bank of Jordan was launched an electronic servers support the service administrations offering to clients.
program so-called (E-fawateercom) since 2015 [2]. This In turn, the (SWAN) provides secured as well as high-speed
program has some important benefits such as reduce the time, connectivity for Government operation between connecting
money and effort of paying bills compared to the traditional State Headquarters and District Headquarters. The e-
way. government framework bolsters correspondence with different
servers, for example, Internet servers to support the service
Government aspects have also changed, including the administrations offering to clients. In turn, the (SWAN)
departments by providing public service oriented to satisfy provides secured and high-speed connectivity for Government
most people [3]. As a result, these Changes in the general functioning between connecting State Headquarters and
society division are described by the improvement of District Headquarters.
electronic government or supposed e-government. In general,
this study suggests dimensions of the most critical aspects
towards the individual’s choice to utilize e-government
services in Jordan.
55 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
56 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A pre-structured interview was done to assess the Based on the employees' opinion, the project of e-
dependent variable that is going to be analyzed is the government needs a short period may not exceed two years at
continuity of using e-government. This was conducted in a their discretion.
pleasant atmosphere. Other independent variables are going to
be analyzed to see its effect on the dependent variable. These B. Do you Encourage the use of e-Government? And why?
dependent variables are the government readiness to apply e- All of the participants encouraged the use of e-
government for a hand, and on the other hand the citizen government; each one gave his/her perspective depending on
willingness to utilize e-government application with the cons the nature of his/ her employment.
and pros of using this application on both parties, the The employee in health department suggesting the issue of
government and the citizens. honesty as he declared that:
All questions were concerned with the e-government "The access of data will be easier and the ability of
issues, like its benefits and online information privacy and manipulation of the data and stocks of medicines and other
security. Furthermore, the interviewees were also stimulated gadgets will be limited. Everything will be present before each
to make comparisons between online as well as offline single person in the department, so the censorship will be
government convenience. The strength of this research is that more intense. Each employee will be doing his/her job
it provides detailed evaluation of the resources, services, and sincerely, and each patient will take his/her right in having the
agency. Also, it backs strategy formulation at agency level. most proper health care."
Conversely, it does not offer details required for effective
strategy formulation at different levels of governance like One of the participants from the social security, however,
community and local levels, of multi-level governance. had her doubts about the application of e-government as she
remembered some bad experiences with the connection to the
V. RESULT AND DISCUSSION web:
Assessing 16 participants did the research from different "Sometimes I feel that it is a good step forward, until the
governmental sectors as it was mentioned in the methodology. connection with the network is disconnected, or the system
However, the sample size of the participants was small but when it gets down and transactions are stopped until
adequate for qualitative methods, since the primary concern of reconstruction is done, I feel that this is a quiet endangering.
this research is placed on the deriving of considerable data However, we all hope that every bad and good expectation is
rather than the verification of hypotheses [11].The textual given into consideration and a sort of pro-activeness is
outcomes of the focus group interviews were content assessed applied to overcome a bad circumstance before it occurs."
by the authors of this study.
Moreover, saving time with a high level of accuracy and
All the differences were solved through meeting to capture less chances to do mistakes were the points, which were
significant facets of e-government aspects. The interview was repeated by the rest of the participants.
constituted from five questions that assessed the dependent
and the independent factors; they are going to be discussed as C. Do you Think that the Jordanian Society can Adopt e-
follows. Government in their Transactions?
A. When do you Expect to Fully Rely on e-Government and This question had variations in answers among different
the Traditional Paper Works Omitted? participants, some of them said that it is possible; others said
that it is possible but some obstacles might face the users.
All of the interviewees had the opinion that fully turning
toward e-government will be soon, in line with the Jordanian Different economic levels are present in the Jordanian
Government's policy towards digitizing the transactions. Society, and the vast majority of them are classified as a low
Some of them had the opinion that the e-government process economic level families and members.
had already started as the employee in the traffic department,
From the perspective of an employee works in health
municipality, tele-communication companies and the Jordan
stated that adopting e-government might be not easy to the
electricity have stated:
Jordanian society. While employee in the Jordanian Electricity
"Paper works in our field of work cannot be omitted, Company and municipality focused on the idea of change and
however, the application of e-government has started to be the presence of willing on the society to adopt new methods,
applied and a website with usernames is being used for with the presence of doubts and fears to use these new
different services such as tax records and invoice numbers." methods as she stated:
While another participant stated that it needs more time, as "Change is hard, and the Jordanian Society is known to be
the employee of health department stated: reluctant to any new practice that would make them adopt new
methods and train on the way of using them, even if it would
"As it is obvious from the updates to the system, and the make their lives easier.
new applied program of visits which is under the test now, I
expect that within one year, and maximum for two years. In addition, the culture of doubt is present in most of the
Paperwork will be limited and the whole information will be members of the society; they will be hardly convinced that this
saved on the computers such as the medications exist at the step will be for their own benefit. They will be worried about
pharmacies, the medical prescriptions of the patients, and their safety and privacy, and convincing them will be hard."
booking time with their doctors, etc."
57 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
From the side of tele-communication companies and well as utilization of government Internet benefits in
traffic department were optimistic and believed that this Netherland [14], a research focused on the client who uses e-
awareness is spread to the members of the society. They also government in Belgium [15], and assessment of government
added: e-charge sites in Sweden [16].
"They can adopt e-government services; however, they will Prominently, only a couple of research examined the
need to have an e-fawateercom account which needs to have a selection in developing nations, for example, learn regarding
bank record which might be hard on some citizens" e-government reception in Cambodia [17], and surveying
resident appropriation of e-government activities in Gambia
D. Do you Think that the Government has the Appropriate [18]. In outcome, little consideration was gotten to analyze e-
Infrastructure for such a Change? government appropriation and utilize in the emergent nation as
Most of them had the opinion that the government a rule.
infrastructure is appropriate as a starting experience, however,
it needs to be updated and more training to the employees is Furthermore, [19] had focused on the application of e-
required. While one of the participants in the traffic government in Saudi Arabia. The qualitative research done
department was very pleased regarding the government and a survey were both done to find the presence of success
infrastructure, however, repairs are required based on the factors such as presence of software and technology that
users' feedback as he stated: support using e-government, presence of customer service, the
level of education as well as computer skills present upon
"Yes and new gadgets are being spread to policemen to be users, the extent of access between users, the level of privacy
directly linked to the net of traffic and vehicle departments' presented to the users in addition to security, factors related to
database. Based on this, we can say that the infrastructure is religion and culture, financial ability, and other factors which
almost ready. He added that based on the feedback from users might include age and gender.
is the infrastructure needs to be updated continuously."
The findings of this research had shown that the issues of
On the other hand, the view of employee in e-government application success are present, and it is
Communication Company was completely satisfied with the showing a bright future, as there is an extent of acceptance
services of the e-government and the infrastructure of it as he among the Saudi population.
stated:
Parallel with the study of [20] had focused on the opinion
"Yes, the system is working dynamically to be updated, to of e-government upon Jordanian citizens, and it was found
solve the problems that face users of the society. In addition, that the perception of using e-government in Jordan is well
the government is trying to hire qualified employees with a due to seeing helpfulness, usability, social impact, respond,
good experience and a wide background in IT, so the hitches and similarity, yet the cost of this service is one of the
are going to be taken into consideration in advance." suppression aspects that negatively affect using e-government.
Another study was done earlier by [21], had assessed the
E. What are the Factors that Affect e-Government Adoption?
satisfaction of the users using an electronic survey. The study
After the analysis the factors can be separated into two demonstrated the significance to uncover the key drivers of e-
groups, factors related to the government, and other factors fulfillment in order to give criticism in many proposals that
related to the citizen. will empower making e-Government gateway which are
The factors related to the government are dependent on the perfect with the people' needs and desires.
IT infrastructure and the employees' knowledge and
TABLE. I. DIMENSIONS OF E-GOVERNMENT
willingness to follow-up and update it. While the factors
related to the citizen are related to the economic situation of Dimensions Description
the citizen as well as the background and the culture of the
families. Check-out process
Payment method
Table I summarizes forth the major dimensions of the Perceived Usefulness
Reducing cost
critical facets of e-government derived from the results, and Saving time
Saving energy
their associated descriptions. Useful information
Compared to our view of results, past researches about the Easy navigation
selection of e-government has examined a few considers, for Perceived Ease to Use Accessible anywhere
example, trust government, trust in web innovation, use of e- Accessible anytime
government value-based administration and others as a major
Trust on Web Personal data security
issue in most recent five years [3,12,13]. In view of the related
works over, the creators supported that those variables have Word of mouth
generally proposed superior information on various sorts of e- T.V.
Social Influences
Radio
government benefit from a few points of view. Social media
In the meantime, previous research on e-government Internet coverage
appropriation has for the most part centered on the developed IT Infrastructure Computers and Tablets
nations; for example, learn concerning the acknowledgment as Websites design
58 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
VI. CONCLUSION [7] Zejnullahu F, Baholli I. Overview of researches on the influential factors
of m-government's adoption. European Journal of Management and
This study proposes dimensions of the most critical factors Marketing Studies. 2017 Aug 11.
towards the individual’s choice to utilize e-government [8] Moon M. J., Welch, E., W. & Wong, W. 2005. What Drives Global E-
services in Jordan. As so far, Jordan is going to follow the Governance? An Exploratory Study at a Macro Level. InProceedings of
trend of adopting e-government, as it has positive effects on the 38th Hawaii International Conference on System Sciences.
https://www.researchgate.net/publication/221179057_What_Drives_Glo
the quality of daily services transactions. In addition, bal_EGovernance_An_Exploratory_Study_at_a_Macro_Level Date of
spreading the awareness among the society members will access 2017 Mar (Vol. 12).
encourage them to use this application. [9] Majdalawi YK, Almarabeh T, Mohammad H, Quteshate W. E-
government strategy and plans in Jordan. Journal of Software
In the end, the concept of e-government is constantly Engineering and Applications. 2015 Apr 14;8(04):211.
developing and changing to keep abreast of technological
[10] Witarsyah D, Sjafrizal T, Fudzee MD, Farhan M, Salamat MA. The
changes in term of designing and promoting services. In this critical factors affecting e-government adoption in Indonesia: A
context, there is an urgent necessity to find creative methods conceptual framework. International Journal on Advanced Science,
of evolving lasting beneficial services to fill gaps between Engineering and Information Technology. 2017 Feb 22;7(1):160-7.
service performance and people anticipations particularly in [11] Izogo EE, Jayawardhena C. Online shopping experience in an emerging
developing countries. e‐retailing market: Towards a conceptual model. Journal of consumer
Behaviour. 2018 Jul;17(4):379-92.
VII. FUTURE STUDIES [12] Alam MZ, Hu W, Barua Z. Using the UTAUT model to determine
factors affecting acceptance and use of mobile health (mHealth) services
The study needs to include more participants from in Bangladesh. Journal of Studies in Social Sciences. 2018 Dec 1;17(2).
different government sectors; in addition, it would give [13] AL-Hujran, O., AL-Debei, M. M., Chatfield, A. and Migdadi, M. The
additional conclusions when citizens are included in the study. imperative of influencing citizen attitude toward e-government adoption
This study is not free of limitations, such as a qualitative and use. Computers in human Behavior,53(1), 2015,189-203.
technique is used in the study. Future studies therefore could [14] Wirtz BW, Piehler R, Rieger V, Daiser P. E-government portal
use a quantitative study in seeking respondents’ opinions. To information performance and the role of local community interest.
Empirical support for a model of citizen perceptions. Public
put it plainly, A TAF model could be adopted to examine the Administration Quarterly. 2016 Apr 1:48-83.
most important factors affecting people's perception towards
[15] Wirtz BW, Kurtz OT. Local e-government and user satisfaction with
acceptance of e-government in Jordan. city portals–the citizens’ service preference perspective. International
REFERENCES Review on Public and Nonprofit Marketing. 2016 Oct 1;13(3):265-87.
[1] Laith T. Khrais. Toward A Model For Examining The Technology [16] Arias MI, Maçada AC. Digital Government for E-Government Service
Acceptance Factors In Utilization The Online Shopping System Within Quality: a Literature Review. InProceedings of the 11th International
An Emerging Markets, 9 (11), 2018, 1099- 1110. Conference on Theory and Practice of Electronic Governance 2018 Apr
4 (pp. 7-17). ACM.
[2] Laith T. Khrais. The Impact Dimensions of Service Quality on the
AcceptanceUsage of Internet Banking Information Systems, American [17] Rana NP, Dwivedi YK, Williams MD. A meta-analysis of existing
Journal of applied sciences, 15(4), 2018, 240-250. research on citizen adoption of e-government. Information Systems
Frontiers. 2015 Jun 1;17(3):547-63.
[3] Alzahrani L, Al-Karaghouli W, Weerakkody V. Analysing the critical
factors influencing trust in e-government adoption from citizens’ [18] Jung D. " Assessing citizen adoption of e-government initiatives in
perspective: A systematic review and a conceptual framework. Gambia: A validation of the technology acceptance model in
International Business Review. 2017 Feb 1;26(1):164-75. information systems success". A critical article review, with questions to
its publishers. Government Information Quarterly. 2019 Jan 1;36(1):5-7.
[4] Kacem A, Belkaroui R, Jemal D, Ghorbel H, Faiz R, Abid IH. Towards
improving e-government services using social media-based citizen's [19] Basahel, A. and Yamin, M. Measuring success of e-government of Saudi
profile investigation. InProceedings of the 9th International Conference Arabia. International Journal of Information Technology, 9(1), 2017,
on Theory and Practice of Electronic Governance 2016, 187-190. ACM. 287-293.
[5] Hashim HS, Hassan ZB, Hashim AS. Factors influence the adoption of [20] Abu-Shanab, E. and Haider, S. Major factors influencing the adoption of
cloud computing: A comprehensive review. International Journal of m-government in Jordan. Electronic Government. An International
Education and Research. 2015;3(7):295-306. Journal, 11(1), 2015,223-240.
[6] Al-Khatib H, Lee H, Suh C, Weerakkody V. E-government systems [21] Sachan A, Kumar R, Kumar R. Examining the impact of e-government
success and user acceptance in developing countries: The role of service process on user satisfaction. Journal of Global Operations and
perceived support quality. 2019. Strategic Sourcing. 2018 Nov 19;11(3):321-36.
59 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Marija Bezbradica3
School of Computing, Dublin City University, Dublin, Ireland
Abstract—Customer support has become one of the most Artificial intelligence (AI) improves digital marketing in a
important communication tools used by companies to provide number of different areas from banking, retail, and travel to
before and after-sale services to customers. This includes healthcare and education. While the idea of using human
communicating through websites, phones, and social media language to communicate with computers holds merit, AI
platforms such as Twitter. The connection becomes much faster scientists underestimate the complexity of human language, in
and easier with the support of today's technologies. In the field of both comprehension and generation. The challenge for
customer service, companies use virtual agents (Chatbot) to computers is not just understanding the meanings of words, but
provide customer assistance through desktop interfaces. In this understanding expression in how those words are collocated.
research, the main focus will be on the automatic generation of
Moreover, a chatbot is an example of a virtual conversational
conversation “Chat” between a computer and a human by
developing an interactive artificial intelligent agent through the
service robot that can provide human-computer interaction.
use of natural language processing and deep learning techniques Companies use robotic virtual agents (Chatbot) to assist
such as Long Short-Term Memory, Gated Recurrent Units and customers through desktop interfaces [1, 2].
Convolution Neural Network to predict a suitable and automatic Natural language processing (NLP) is a subfield of
response to customers’ queries. Based on the nature of this computer science that employs computational techniques for
project, we need to apply sequence-to-sequence learning, which learning, understanding and producing human language
means mapping a sequence of words representing the query to content. NLP can have multiple goals; it can aid human-human
another sequence of words representing the response. Moreover,
communication, such as in machine translation and aid human-
computational techniques for learning, understanding, and
producing human language content are needed. In order to
machine communication, such as with conversational agents.
achieve this goal, this paper discusses efforts towards data Text mining and natural language processing are widely used
preparation. Then, explain the model design, generate responses, in customer care applications to predict a suitable response to
and apply evaluation metrics such as Bilingual Evaluation customers, which significantly reduces reliance on call center
Understudy and cosine similarity. The experimental results on operations [3].
the three models are very promising, especially with Long Short- AI and NLP have emerged as a new front in IT customer
Term Memory and Gated Recurrent Units. They are useful in
service chatbots. The importance of these applications appears
responses to emotional queries and can provide general,
when no technicians manage the customer service office due to
meaningful responses suitable for customer query. LSTM has
been chosen to be the final model because it gets the best results the end of working time or their presence outside the office [4].
in all evaluation metrics. In this project, the main focus will be on the automatic
generation of conversation ”Chat” between a computer and a
Keywords—Chatbot; deep learning; natural language human by developing an interactive artificial intelligent agent
processing; similarity using deep learning. This will provide customers with the right
I. INTRODUCTION information and response from a trusted source at the right
time as fast as possible.
With the arrival of the information age, customer support
has become one of the most influential tools companies use to This project aims to build an automated response system
communicate with customers. Modern companies opened up (Chatbot) that responds to customer queries on social
communication lines (conversations) with clients to support networking platforms (Twitter) to accelerate the performance
them regarding products before and after-sales through of the service. Also, to keep the simplicity in mind while
websites, telephones, and social media platforms such as designing the system to enhance its efficiency.
Twitter. This communication becomes faster and much easier
with the support of the technologies that are being used today.
60 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
This project centers around the study of deep learning manual intervention and supervision, which affects the speed
models, natural language generation, and the evaluation of the and quality of processes execution. IT service providers are
generated results. under competitive pressure to continually improve their service
quality and reduce operating costs through automation. Hence,
We believe that this contribution can add improvement by they need the adoption of chatbots in order to speed up the
applying the right preprocessing steps which may organize work and ensure its quality [10].
sentences in a better way and help in generating proper
responses. On the other hand, we start with the existing text On the medical side, the field of healthcare has developed a
generative models CNN and LSTM and then try to improve lot, lately. This development appears with the use of
them as well as develop a new model such as GRU to compare information technology and AI in the field. In [11], the authors
results. We focus on evaluating the generated responses from proposed a mobile healthcare application as a chatbot to give a
two aspects: the number of words matches between the fast treatment in response to accidents that may occur in
reference response and the generated response and their everyday life, and also in response to the sudden health
semantic similarity. changes that can affect patients and threaten their lives.
The rest of this paper is organized as follows. Section II Customer services agent is an application of applying
provides reviews of the related works. The methodological chatbot technologies in businesses to solve customer problems
approach is described in Section III. Moreover, dataset and help the sales process. As companies become globalized in
collection and analysis in details are provided in Section IV. the new era of digital marketing and artificial intelligence,
The implementation strategy and results of this project are brands are moving to the online world to enhance the customer
discussed in section V. Finally, the conclusion of the project experience in purchasing and provide new technical support
and its future work are provided in Sections VI and VII ways to solve after-sales problems. Moreover, fashion brands
respectively. such as Burberry, Louis Vuitton, Tommy Hilfiger, Levi's,
H&M, and eBay are increasing the popularity of e-service
II. LITERATURE REVIEW agents [1].
Developing computational conversational models B. Natural Language Processing
(chatbots) took the attention of AI scientists, for a number of
years. Modern intelligent conversational and dialogue systems NLP allows users to communicate with computers in a
draw principles from many disciplines, including philosophy, natural way. The process of understanding natural language
linguistics, computer science, and sociology [5]. This section can be decomposed into the syntactic and semantic analysis.
will explore the previous work of chatbots and their Syntactic refers to the arrangement of words in a sentence such
implementations. that they make grammatical sense. Moreover, syntactic
analysis transforms sequences of words into structures that
A. Chatbots Applications and Uses show how these words are related to each other. On the other
Artificial dialogue systems are interactive talking machines hand, semantic refers to the meaning of each word and
called chatbots. Chatbot applications have been around for a sentence. The semantic analysis of natural language content
long time; the first well-known chatbot is Joseph captures the real meaning; it processes the logical structure of
Weizenbaum‟s Eliza program developed in the early 1960s. sentences to find the similarities between words and
Eliza facilitated the interaction between human and machine understand the topic discussed in the sentences [12].
through a simple pattern matching and a template-based As part of the text mining process, the text needs many
response mechanism to emulate the conversation [6, 7]. modification and cleaning before using it in the prediction
Chatbot became important in many life areas; one of the models. As mentioned in [13], the text needs many
primary uses of chatbots is in education as a question preprocessing steps which include: removing URLs,
answering system for a specific knowledge domain. In [8], the punctuation marks and stop words such as a, most, and, is and
authors proposed a system that has been implemented as a so on in the text because those words do not contain any useful
personal agent to assist students in learning Java programming information. In addition, tokenizing, which is the process of
language. The developed prototype has been evaluated to breaking the text into single words. Moreover, text needs
analyze how users perceive the interaction with the system. stemming, which means changing a word into its root, such as
Also, the student can get help in registering and dropping “happiness” to “happy”. For features extraction, the authors
courses by using a chatbot spatialized in student administrative use Bag of Words (BoW) to convert the text into a set of
problems, as mentioned in [9]. The administrative student‟s features vector in numerical format. BoW is the process of
chatbot helps the colleges to have 24*7 automated query transforming all texts into a dictionary that consist of all words
resolution and helps students have the right information from a in the text paired with their word counts. Vectors are then
trusted source. formed based on the frequency of each word appearing in the
text.
On another hand, information technology (IT) service
management is an important application area for enterprise Before entering the data into a model or a classifier, it is
chatbots. In many originations and companies, IT services desk necessary to make sure that the data are suitable, convenient,
is one of the essential departments that helps to ensure the and free of outliers. In [14], the authors explain how to
continuity of work and solving technical problems that preprocess the text data. The main idea was to simplify the text
employees and clients are facing. This variability demands for the classifier to learn the features quickly. For example, the
61 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
names can be replaced with one feature {{Name}} in the The customer service agent is an important chatbot that is
feature set, instead of having the classifier to learn 100 names used to map conversations from request to the response using
from the text as features. This will help in grouping similar the sequence to sequence model. Moreover, a sequence to
features together to build a better predicting classifier. On sequence models has two networks one work as an encoder
another hand, emoticons and punctuation‟s marks are that maps a variable-length input sequence to a fixed-length
converted to indicators (tags). Moreover, a list of emoticons is vector, and the other work as a decoder that maps the vector to
compiled from online sources and grouped into categories. a variable-length output sequence. In [4], the authors generate
Other punctuation marks that were not relevant to the coding word-embedding features and train word2vec models. They
scheme are removed. trained LSTMs jointly with five layers and 640 memory cells
using stochastic gradient descent for optimization and gradient
Chat language contains many abbreviations and clipping. In order to evaluate the model, the system was
contractions in the form of short forms and acronyms that have compared with actual human agents responses and the
to be expanded. Short forms are shorter representations of a similarity measured by human judgments and an automatic
word which are done by omitting or replacing few characters, evaluation metric BLEU.
e.g., grp → group and can‟t → cannot. The authors created a
dictionary of these words from the Urban Dictionary to replace As a conclusion of reviewing works concerned with the
abbreviations by expansions. Spell checking is performed as conversational system, text generation in English language and
the next step of the pre-processing pipeline on all word tokens, the collaboration of social media in customer support service,
excluding the tagged ones from the previous steps [14]. this paper proposes a work that aims to fill the gap of limited
works in the conversational system for customer support field,
Minimizing the words during the text pre-processing phase especially in the Twitter environment. The hypothesis of this
as much as possible is very important to group similar features project was aiming to improve the automated responses
and obtain a better prediction. As mentioned in [15], the generated by different deep learning algorithms such as LSTM,
authors suggest processing the text through stemming and CNN, and GRU to compare results and then evaluate them
lower casing of words to reduce inflectional forms and using BLEU and cosine similarity techniques. As a result, this
derivational affixes from the text. The Porter Stemming project will help to improve the text generation process in
algorithm is used to map variations of words (e.g., run, running general, and customer support field in particular.
and runner) into a common root term (e.g., run).
Words can not be used directly as inputs in machine III. METHODOLOGICAL APPROACH
learning models; each word needs to be converted into a vector This section discusses the background of the implemented
feature. In [4], the authors adopt the Word2vec word methods, explain why these methods are appropriate and give
embedding method to learn word representations of customer an overview of the project methodology.
service conversations. Word2vec's idea is that each dimension
A. Text Generative Model
of inclusion is a possible feature of the word, which can
capture useful grammatical and semantic properties. Moreover, Based on the nature of this project, which is generating a
they tokenize the data by building a vocabulary of the most proper response to every customer query in social media,
frequent 100K words in the conversations. applying sequence-to-sequence learning are needed. Moreover,
sequence-to-sequence means mapping a sequence of words
C. Machine Learning Algorithm and Evaluation representing the query to another sequence of words
A large number of researchers use the idea of artificial representing the response, the length of queries and responses
intelligence and deep learning techniques to develop chatbots can be different. This can be applied by the use of NLP and
with different algorithms and methods. As mentioned in [16], deep learning techniques.
the authors use a repository of predefined responses and a
Sequence-to-sequence models are used in many fields,
model that ranks these responses to pick an appropriate
including chat generation, text translation, speech recognition,
response for a user‟s input. Besides, they proposed topic aware
and video captioning. As shown in Fig. 1, a sequence-to-
convolutional neural tensor network (TACNTN) model to
sequence model consists of two networks, encoder, and
classify whether or not a response is proper for a message. The
decoder. The input text enters the encoder network in reverse
matching model used to select a response for a user message.
order, then it is converted into a sequence of fixed length
Specifically, it has three-stages that include: pre-processing the
context vector, which is then used by the decoder to generate
message, retrieving response candidates from the pre-defined
the output sequence [18].
message-response pair index, then ranking the response
candidates with a pre-train matching model.
In [17], the authors train two word-based machine learning
models, a convolutional neural network (CNN) and a bag of
words SVM classifier. Resulting scores are measured by the
Explanatory Power Index (EPI). EPI used to determine how
much words contribute to the classification decision and filter
relevant information without an explicit semantic information
extraction step. Fig. 1. Sequence to Sequence Model.
62 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
63 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. DATASET FEATURES DESCRIPTION As shown in the above figure, around ten brands take more
than two days (60 hours) to respond to customers queries,
Feature Description Datatypes which may cause problems to customers, effect companies'
reputation and the customers may start looking for other
A unique, anonymized ID for the
Tweet. Referenced by
service providers.
tweet_id int64
response_tweet_id and A filtering process is used to convert the dataset records
in_response_to_tweet_id.
into a conversational dataset suitable for the experiments. The
A unique, anonymized user ID. The
filtering is done as follows:
real user_id in the dataset has been
author_id
replaced with their associated
object 1) Pick only inbound tweets that are not in reply to any
anonymized user ID. other tweet.
2) Organize each tweet with the corresponding reply by
Whether the tweet is "inbound" to a matching in_response_to_tweet_id with tweet_id features.
company doing customer support
3) Filter out cases where reply tweets are not from a
Inbound on Twitter. This feature is useful bool
when reorganizing data for training company based on the in inbound feature (if the inbound
conversational models. feature is False it means that the tweet is from a company;
otherwise it is from a user).
Date and time when the tweet was
created_at object
sent. However, when revising the dataset, it has been found that
some of the tweets have no replies at all; they are from multiple
Tweet content. Sensitive languages, and some of them are just samples and emojis. For
information like phone numbers
Text
and email addresses are replaced
object this type of tweets further preprocessing step is performed to
with mask values like __email__. remove non-English tweets by the use of the langdetect library
which detects any non-English text [24]. Then, the non-
IDs of tweets that are responses to responses English tweets are studied, as shown in the word
response_tweet_id object
this tweet, comma-separated. cloud in Fig. 7, (which is a graph that illustrates the most words
that appear in the text).
ID of the tweet this tweet is in
in_response_to_tweet_id float64
response to, if any. It can be observed that the words appear with no hint to a
specific problem discussed, and most of the queries are
thanking the customer support services for example:
@AmazonHelp Thanks for the quick response
@AppleSupport Awesome, thanks
Others asking for help in general:
@Uber_Support Sent a DM Hope you could help soon.
@O2 DM sent. Still no further forward!
The modified dataset contains 794,299 rows and 6 columns
which are: author_id_x, created_at_x, text_x, author_id_y,
created_at_y and text_y. X refers to the queries, and Y refers to
the responses from customer support teams.
Fig. 6. The Average Delay in Response to Customers in Hours per Brand. Fig. 7. Most Words used in the Queries without Responses Data.
64 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
V. IMPLEMENTATION STRATEGY TABLE. II. THE CHANGES IN TEXT BEFORE AND AFTER APPLYING
PREPROCESSING STEPS
In this section, we are going to explain the methodology
followed for this project. At first, prepare the dataset for Before preprocessing After preprocessing
modeling. The preparing process includes preprocessing step
and features extraction then train the models using a training @115743 C91. Feel free to keep an
feel free to keep an eye on the ps
set and evaluate them with a test set. eye on the PS Blog for news and
blog for news and updates
updates: https://t.co/aLtfBAztyC
A. Data Preprocessing
A data analyst cannot handle raw text directly to suit @133100 We do our best to clear as we do our best to clear as many
machine learning or deep learning methods. Therefore, it is many upgrades as we can, send us a upgrades as we can send us a dm
necessary to work on texts‟ preprocessing from all existing DM with the reservation you're with the reservation you are referring
referring to and we'll take a look. to and we will take a look
impurities, for example, punctuation, expression code, and
non-English words (Chinese, Spanish, French, and others). In
@129388 We'd like to look into this
order to do this, a number of python NLP libraries such as with you. To confirm, did you update
we would like to look into this with
regular expression (RE), unicodedata, langdetect, and you to confirm did you update to ios
to iOS 11.1? Please DM us here:
contractions are used. please dm us here
https://t.co/GDrqU22YpT
In this project, the performed preprocessing steps include:
remove links, images, Twitter ID, numbers, punctuation, emoji, TABLE. III. TRAINING TIME IN HOURS
non-English words and replace abbreviations with long forms.
Table II illustrates the changes in the dataset before and after Model Training Time in Hours
applying all the previous preprocessing steps. LSTM 12
The preprocessing steps are chosen carefully; not all GRU 8
preprocessing techniques are suitable for this kind of projects. CNN 5
For example, removing stopwords and text stemming cannot
be applied because it will affect the sentences structures as well In the experiments, multiple parameters are tested, and their
as the text generation process. effects are addressed. All models are tested with varying
B. Feature Extraction dimensionality of the word embeddings (100, 300 and 640), it
was observed that models perform better and faster with 100-
Before doing any complex modeling, the dataset needs to word embedding size.
be transformed into a numerical format suitable for training.
The Bag of Words (BOW) concept is applied to extract The dataset is large, the number of vocabularies is 388,950
features from the text dataset. First, all of the texts in the unique words, and our computers cannot handle it. So, only the
dataset are split into an array of tokens (words). Then, a frequent words appeared in the dataset should be used. The
vocabulary dictionary is built with all of the words in the most frequent words are decided by the max_features
dataset and its corresponding index value. The array of words parameter in the CountVectorizer function which sort words by
is then converted to an array of indexes. This process can be its frequency then choose the most frequent words. The first
applied by the use of the sklearn‟ predefined method called vocabulary size in the experiments is 8000 and then it
CountVectorizer. increases, taking into consideration memory limitation. A
slight improvement has been recognized in all models and
In order to handle variable length, the maximum sentence because of the memory limitation, only 10,000 of the
length needs to be decided. Moreover, all remaining vector vocabularies are used. Moreover, the GRU model was trained
positions should be filled with a value („1‟ in this case) to make for eight epochs but without significant improvement. The
all sequences have the same length. On the other hand, words three models are all trained under the same conditions. Table
not in the vocabulary dictionary will be represented with UNK IV shows the common parameters used in all models.
as a shortcut of unknown words. Moreover, each output text in
the dataset will start with a start flag („2‟ in this case) to help in TABLE. IV. THE COMMON PARAMETERS USED IN LSTM, GRU AND CNN
training. Now the dataset is ready for training. MODELS
65 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The following are the common layers used in the models, Calculate brevity penalty (equation 1): a penalization is
starting from inserting the sequence of words into the model to applied to short answers, which might be incomplete.
generating the responses:
Last Word Input Layer: Inputs the last word of the BP={ ( ( )) (1)
sequence.
, where c is the length of generated response and r is the
Encoder Input Layer: Inputs sequence data and pass it to length of reference response.
the embedding layer.
Then, calculate the BLEU score (equation 2) [23]:
Embedding Layer: Used to create word vectors for ∑ ( ( ))
incoming words. BLEU (2)
Encoder Layer (LSTM, GRU, CNN): Creates a , where Wn = 1/N.
temporary output vector from the input sequence. 2) Cosine Similarity: On the other hand, cosine similarity
Repeated Vector Layer: Used like an adapter to fit the also used to compute the similarity between the generated
encoder and decoder parts of the network together. It can response and the reference response in vector representation. If
be configured to repeat the fixed-length vector one time there is more similarity between the two vectors, the cosine
for each time step in the output sequence. similarity value is near to one; otherwise, it is near to zero.
Concatenate Layer: Takes inputs and concatenates them 3) In order to implement the cosine similarity, the pre-
along a specified dimension. trained model word2vec are used. The word2vec model is in
gensim package, and it has been trained on part of Google
Decoder Layer (LSTM, GRU, CNN)(Dense): Used as News dataset (about 100 billion words) [25]. The model
the output for the network.
contains 300-dimensional vectors for 3 million words and
Next Word Dense Layer: Takes inputs from the previous phrases.
layer and outputs a one vector representing the target
The word2vec model used to represent words in a vector
word.
space [26]. Words are represented in the form of vectors and
Next Word softmax Layer: Applies a softmax function placement is done in such a way that similar meaning words
that turns the dense layer output into a probability appear together, and different words are located far away.
distribution, from to pick the most likely next word.
D. Generating Responses
After training the models, the generating responses process
is started using the 30k test set. The following are samples of
the generated responses from all models (see Fig. 8 and 9).
E. Evaluation
The Bilingual Evaluation Understudy and cosine similarity
evaluation metrics are used to compute the similarity between
the generated response and the reference response.
1) Bilingual Evaluation Understudy (BLEU): BLEU was
originally created to measure the quality of machine translation
with respect to human translation. It calculates an n-gram
precision (An n-gram is a sequence of n words that appear Fig. 8. Good Result Example.
consecutively in the text) between the two sequences and also
imposes a commensurate penalty for machine sequence being
shorter than human one. A perfect match score is 1.0, whereas
a perfect mismatch score is 0.0.
The computation of BLEU involves various components:
n-gram precisions (Pn) and BLEU‟s brevity penalty. Those
measures are calculated as shown in the following steps:
Calculate n-gram precision (Pn): measures the frequency
of the n-gram according to the number of times it
appears in the generated response and reference
response. Pn must be calculated for each value of n,
which usually ranges from 1 to 4. Then the geometric
average of Pn should be computed with a weighted sum
of the logarithms of Pn. Fig. 9. Bad Result Example.
66 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Gensim is a topic modeling toolkit which is implemented in Furthermore, another evaluation metric cosine similarity
python. Topic modeling is discovering the hidden structure in are applied to captures the semantics beyond responses and
the text body. Word2vec model is imported from Gensim gives similarity scores. It has been found that RNN models
toolkit and uses a built-in function to calculate the similarity capture the semantics in the responses and they are more
between the generated response and reference response. effective in improving the reply quality than the CNN model.
Fig. 11 shows the similarity scores for each model.
F. Result and Discussion
Before discussing and reviewing the results, the most After exploring the generated responses and get in-depth in
important features of the baseline model are discovered to have the good and bad results, it has been found that RNN models,
a rich discussion with clear comparisons. Table V shows the in general, are good in responses to emotional queries more
baseline model implementation. than an informative one. The models can provide general,
meaningful responses suitable for customer query. Table VI
In this project, the process of generating responses take shows an example of an emotional query.
around 6 hours for each model to be accomplished. Moreover,
calculating BLEU and cosine similarity scores takes around 4 On the other hand, the queries that are more informative
hours. and ask about specific information are hard to generate, and the
generated responses become less efficient. Table VII shows an
The models are evaluated automatically based on the words example of an informative query.
using BLEU score. The BLEU is applied for 1-gram, 2-gram,
3-gram, and 4-gram in order to explore the strength of the By looking at the different responses from different models,
models. It can be seen that LSTM and GRU models it has been noticed that LSTM is generating better sentences
outperform the official baseline LSTM model [4] with respect that make sense and it is hard to say if the response is from a
to the 4-gram BLEU score. Fig. 10, shows in details the human or machine whereas GRU responses are not as good as
performance of models in each n-gram. LSTM.
Optimization Function Stochastic gradient descent and gradient clipping. TABLE. VI. EXAMPLE OF EMOTIONAL QUERY AND RESPONSES FROM ALL
Evaluation BLEU with the best score achieved 0.36. MODELS
67 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. VII. EXAMPLE OF INFORMATIVE QUERY AND RESPONSES FROM [5] S. Oraby, P. Gundecha, J. Mahmud, M. Bhuiyan, and R. Akkiraju,
ALL MODELS “Modeling Twitter Customer ServiceConversations Using Fine-Grained
Dialogue Acts,” in Proceedings of the 22nd International Conference on
guys when are you going to open your services Intelligent User Interfaces - IUI ‟17, 2017, pp. 343–355.
Customer Query
in middle east
[6] H. Shah, K. Warwick, J. Vallverdú, and D. Wu, “Can machines talk?
hulu store is only available in the us at this time Comparison of Eliza with modern dialogue systems,” Comput. Human
Customer Support Behav., vol. 58, pp. 278–295, May 2016.
but we will share the interest in bringing our
Response
service to the middle east [7] R. DALE, “The return of the chatbots,” Nat. Lang. Eng., vol. 22, no. 05,
pp. 811–817, Sep. 2016.
hi there we are sorry to hear about this please
LSTM Generated [8] M. Coronado, C. A. Iglesias, Á. Carrera, and A. Mardomingo, “A
dm us with your email address so we can
Response cognitive assistant for learning java featuring social dialogue,” Int. J.
connect Hum. Comput. Stud., vol. 117, pp. 55–67, Sep. 2018.
hi there i am sorry to hear about this please dm [9] S. Jha, S. Bagaria, L. Karthikey, U. Satsangi, and S. Thota, “STUDENT
GRU Generated INFORMATION AI CHATBOT,” in International Journal of Advanced
me the details of the issue you are having with
Response Research in Computer Science, 2018, vol. 9, no. 3.
your services
[10] P. R. Telang, A. K. Kalia, M. Vukovic, R. Pandita, and M. P. Singh, “A
CNN Generated Conceptual Framework for Engineering Chatbots,” IEEE Internet
hi there is not have you are you
Response Comput., vol. 22, no. 6, pp. 54–59, Nov. 2018.
[11] K. Chung and R. C. Park, “Chatbot-based heathcare service with a
VI. CONCLUSION knowledge base for cloud computing,” Cluster Comput., pp. 1–13, Mar.
2018.
In this project, we build customer support chatbot that helps
[12] J. Savage et al., “Semantic reasoning in service robots using expert
companies to have 24 hours of automated responses. After systems,” Rob. Auton. Syst., vol. 114, pp. 77–92, Apr. 2019.
analyzing the dataset and understanding the importance to have [13] S. T. Indra, L. Wikarsa, and R. Turang, “Using logistic regression
automated responses to customers and companies, we start method to classify tweets into the selected topics,” in 2016 International
exploring existing techniques used for generating responses in Conference on Advanced Computer Science and Information Systems
the customer service field. Then, we attempt to try three (ICACSIS), 2016, pp. 385–390.
different models LSTM, GRU, and CNN. The experimental [14] A. Shibani, E. Koh, V. Lai, and K. J. Shim, “Assessing the Language of
results show that LSTM and GRU models(with modified Chat for Teamwork Dialogue,” 2017.
parameters) tend to generate more informative and valuable [15] A. Singh and C. S. Tucker, “A machine learning approach to product
review disambiguation based on function, form and behavior
responses compared to CNN model and the baseline model classification,” Decis. Support Syst., vol. 97, pp. 81–91, May 2017.
LSTM. Besides, we used a BLEU score and cosine similarity
[16] Y. Wu, Z. Li, W. Wu, and M. Zhou, “Response selection with topic
as evaluation measures to support the final decision. clues for retrieval-based chatbots,” Neurocomputing, vol. 316, pp. 251–
261, Nov. 2018.
VII. FUTURE WORK [17] L. Arras, F. Horn, G. Montavon, K.-R. Müller, and W. Samek,
In future work, we plan to incorporate other similarity “"What is relevant in a text document?": An interpretable
measures such as soft cosine similarity. Also, we plan to machine learning approach,” PLoS One, vol. 12, no. 8, p. e0181142,
Aug. 2017.
improve the experiments by increase the vocabulary size and
[18] S. Sen and A. Raghunathan, “Approximate Computing for Long Short
try to increase the epoch parameters to reach 100 after Term Memory (LSTM) Neural Networks,” IEEE Trans. Comput. Des.
providing proper infrastructure. We further can add more data Integr. Circuits Syst., vol. 37, no. 11, pp. 2266–2276, Nov. 2018.
for the training by taking benefits from the queries without [19] Z. Wang, Z. Wang, Y. Long, J. Wang, Z. Xu, and B. Wang, “Enhancing
responses and translate non-English queries. generative conversational service agents with dialog history and external
knowledge I,” 2019.
ACKNOWLEDGMENT [20] J. Zhang and C. Zong, “Deep Neural Networks in Machine Translation:
This research was funded by the Deanship of Scientific An Overview,” IEEE Intell. Syst., vol. 30, no. 5, pp. 16–25, Sep. 2015.
Research at Princess Nourah bint Abdulrahman University [21] R. C. Gunasekara, D. Nahamoo, L. C. Polymenakos, D. E. Ciaurri, J.
Ganhotra, and K. P. Fadnis, “Quantized Dialog – A general approach for
through the Fast-track Research Funding Program. conversational systems,” Comput. Speech Lang., vol. 54, pp. 17–30,
REFERENCES Mar. 2019.
[1] M. Chung, E. Ko, H. Joung, and S. J. Kim, “Chatbot e-service and [22] G. Aalipour, P. Kumar, S. Aditham, T. Nguyen, and A. Sood,
customer satisfaction regarding luxury brands,” J. Bus. Res., Nov. 2018. “Applications of Sequence to Sequence Models for Technical Support
Automation,” in 2018 IEEE International Conference on Big Data (Big
[2] J. Hill, W. Ford, I. F.-C. in H. Behavior, and undefined 2015, “Real Data), 2018, pp. 4861–4869.
conversations with artificial intelligence: A comparison between
human–human online conversations and human–chatbot conversations,” [23] J. Singh and Y. Sharma, “Encoder-Decoder Architectures for Generating
Elsevier. Questions,” Procedia Comput. Sci., vol. 132, pp. 1041–1048, 2018.
[3] J. Hirschberg and C. D. Manning, “Advances in natural language [24] N. Shuyo, “Language Detection Library for Java.” 2010.
processing,” Science (80-. )., vol. 349, no. 6245, pp. 261–266, Jul. 2015. [25] R. Rehurek and P. Sojka, “Software Framework for Topic Modelling
[4] A. Xu, Z. Liu, Y. Guo, V. Sinha, and R. Akkiraju, “A New Chatbot for with Large Corpora,” in Proceedings of the LREC 2010 Workshop on
Customer Service on Social Media,” in Proceedings of the 2017 CHI New Challenges for NLP Frameworks, 2010, pp. 45–50.
Conference on Human Factors in Computing Systems - CHI ‟17, 2017, [26] Y. Zhu, E. Yan, and F. Wang, “Semantic relatedness and similarity of
pp. 3506–3510. biomedical terms: examining the effects of recency, size, and section of
biomedical publications on the performance of word2vec.,” BMC Med.
Inform. Decis. Mak., vol. 17, no. 1, p. 95, Jul. 2017.
68 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Most of the organization uses information system There are many software quality models currently being
to manage the information and provide better decision making in used to evaluate quality of software products. So, this paper
order to deliver high quality services. Due to that the information will firstly show the comparisons between these models in the
system must be reliable and fulfill the quality aspect in order to literature review section. The advantages and disadvantages of
accommodate organization’s need. However, some of the these models are also investigated. Based on these analyses,
information system still facing problems such as slow response the ISO 25010 model was adopted in this study. Next, the
time, problem with accessibility and compatibility issues between paper discusses the methodology used in the study, which
hardware and software. These problems will affect the includes the descriptions of constructs and the explanation of
acceptance and usage of the information system especially for
the Rasch model used to perform data analysis. The paper also
non-computing users. Therefore, this study was aimed to
investigate the factors that significantly contribute to the quality
shows and discusses findings based on main assumptions of
of software for information system. A survey was carried out by the Rasch Model for selection of items such as item fit,
distributing a set of questionnaires to 174 respondents who are unidimensionality and local independence.
involved in development of software for information system. The The study is very important in that it enables the
data was analyzed using Rasch Measurement Model since it improvement towards the information system development by
provides reliability of respondents and instruments. The result having a guideline on factors that significantly improve the
indicates that 30 factors had significantly contributed to the software quality aspect. It also serves as an additional
quality of software for information system and of these, six
reference towards the improvement of software quality in
factors are under functionality, five for reliability, ten for
usability, five for efficiency, two for compatibility and two for
information systems.
security. It is hoped that by identifying these factors, system However, this study only focuses on the human resource
developers can seriously consider of enhancing the quality of information system widely used in the planning and
software for information system projects. In future, these factors management of human resource. It also considers the software
can be used to develop an evaluation tool or metrix for quality quality factors from the users’ perspective only because
aspects of software for information system projects. software quality issues are usually related to this perspective.
Keywords—Information system; quality of software; Rasch II. LITERATURE REVIEW
measurement model; evaluation; factors
Various models have been developed to measure software
I. INTRODUCTION quality for information system such as the McCall, Boehm,
Information system is important in developing successful FURPS, Dromey, ISO 9126 and ISO 25010 models. Each
and competitive organizations that can deliver high quality model was developed based on a certain unique principal or
products and services to customers [1,2]. It helps improving concept. These models explain about different aspects of
the flow of information and work processes in organisation, software characteristics [5]. These models can be viewed from
thus can enhance the decision-making processes. Even though a user perspective, a manufacturing perspective or a product
the information system provides benefits to the organization, it perspective. Table I below shows the comparison between
has been highlighted some weaknesses that contributed to the these models.
failure of fulfilling the quality aspects of a system such as These models also have the advantages and disadvantages
slow response time, access problem, difficulties in using of their own as stated in Table II below.
system, unavailability and incompatibility between hardware
and software [3,4]. A. Factors Influencing Software Quality
This study also analyzed previous studies to identify the
As a result, if these problems continuously faced by the
quality dimensions and factors that were used to measure the
users, it will cause less acceptance and usage of information
software quality. Table III shows the type of information
systems. Thus, it leads to the poor delivery of services and
systems that were analyzed.
products and finally will damage the organization’s reputation.
Therefore, system developers should focus to strengthen the
software quality aspect of information system.
69 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
70 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
71 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Based on the analysis, it was found that various quality As stated above, the constructs and items (quality factors)
factors used differently according to the types and functions of are gathered based on the analysis of previous studies related
an information system. Several factors are widely used by to software quality. To ensure that the constructs and items are
researchers such as easy to use, response and reliability. This valid and can be used to collect data effectively, the
may be because these three factors reflect the basic features development of survey is very important. It must be done
required to ensure the quality of the system. Table IV shows systematically to ensure it fulfils the study objectives. After
the list of quality factors. identifying constructs and items, a suitable scale is selected
and the testing of item validity and instrument reliability are
III. METHODOLOGY performed. Expert view is obtained and content validation is
In this study, a quantitative approach was used by also done since they are also important elements in making
conducting a survey to achieve its objectives and questions. sure the instrument is relevant.
A. Participants C. Data Analysis
A total of 174 civil servants in Public Service Department Rasch Model is used to analyze data from the respondents.
(PSD), Putrajaya participated in this study. They comprise of The model refers to an idea, principal, guideline or technique
67 males (39%) and females (61%). They were divided into 2 that enables measurement of the latent trait [6]. It basically
categories of services, where 79 or 45% were in the separates individual capabilities and instrument’s quality. This
Professional and Management category, and 95 (55%) were in model assumes that individual response towards an item is
the Support Services category. only influenced by individual capabilities and item difficulties
[7]. The ability of the Rasch Model as an analytical instrument
B. Instrument is proved by its application in various research areas including
This study employs a self-developed 39-items management and social science. This model prevents
questionnaire consisting of six constructs that represent researchers in social sciences area from making a raw and
quality factors, namely the Functionality, Reliability, blurred observation and undertakes definitive actions with
Usability, Efficiency, Compatibility and Security (Table V). realistic accuracy and clear quality control [8]. In this study,
The scale is 5 points Likert-type, where participants are the WinStep software is used to perform the Rasch analysis.
required to give their response on a Strongly Disagree−
Disagree – Slightly Agree – Agree – Strongly Agree pattern.
72 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
updated information) and both characteristics are important, respondents are competent enough to determine the criteria
both are retained in the actual questionnaire. influencing the software quality of an information system.
D. Reliability Index The difficulty measurement value ranges between +3.12
The Cronbach Alpha index is valued at 0.96. This shows logits and -0.96 logits. The item distribution shows item B7r -
that the instrument is highly reliable. The respondent Software problem affects the system performance as the most
reliability index is 0.93, and item reliability index is 0.98. This difficult to be agreed item while item E1 – All data must be
result indicates that there is enough sample and the instrument integrated with each other as the easiest item to be agreed.
is suitable to measure the respondents’ capabilities and item Based on the analysis, there were 30 items that are under
difficulties. the Meanitem = 0.00, indicating the number of criteria that
E. Separation Index contribute significantly to the software quality of the human
resource information system. Table VI shows the significant
The respondent separation index is 3.67, and the item items.
separation index is 7.70. Index value between three and four
indicates good value while the value more than five indicates The gap in the item distribution map is also examined to
excellent value [6]. This shows that the items can separate the determine whethe.r the items are enough to evaluate the
respondents based on their capabilities and items based on the capabilities of all respondents. The result shows there are gaps
difficulties. between items B6r (2.75 logits) and C8r (2.10 logits) and D2r
(1.56 logits) and C10 (0.75 logits). These gaps indicate that
F. Respondent – Item Distribution Map there are not enough items to measure higher level of
The respondent-item map shows the distribution of items respondents’ capabilities. This is also highlighted by the
based on difficulties with the distribution of respondents’ respondents and item distribution above the line, where the
capabilities along the logits scale (see Fig. 1). number of items is comparatively smaller compared to the
number of employees.
Overall, most respondents are above the Meanitem = 0.00
logits. This indicates that almost all of them can answer the Thus, further study is required to develop more difficult
questionnaire. There are also a larger number of respondents items that can measure respondents with higher capabilities.
above the Meanrespondent = 2.08 logits.
TABLE It also
12.2 KUALITI indicates
SISTEM MAKLUMATthat the ZOU432WS.TXT Apr 1 14:41 2019
INPUT: 174 EMPLOYEES 39 items MEASURED: 174 EMPLOYEES 39 items 5 CATS 3.68.2
--------------------------------------------------------------------------------
73 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
74 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
efficient system should also be capable to maintain its requires different capabilities and technical requirements. So it
performance while being accessed by many users at the same is also recommended that future works should look into
time. For example, in public sector there are information different types of information system. This will further
systems that are developed for the benefit of the whole civil contribute to the improvement of the system and software
servants. In this construct, only item D2r – Takes a long time quality.
to process users’ request such as generating report or
analyzing information that is the least agreeable. This is ACKNOWLEDGMENT
because an information system is designed to help users doing This paper was funded by Fundamentals Research Grant
their work. Therefore, they need a system that would be able Scheme (FRGS), Ministry of Education Malaysia under
to give response to a request and process it within a short project code FRGS/1/2018/ICT04/UKM/02/11.
period of time.
REFERENCES
As for Construct E (Compatibility), both items are most [1] Oprea, M., “MAS_UP-UCT: A multi-agent system for university course
agreed by respondent. This shows that both items are very timetable scheduling,” International Journal of Computers,
important in the development of information system. The Communications & Control, 1, 2007, 94–102. doi:10.1590/S1415-
65552003000100014.
capability of data and system integration with different
[2] Masrek, M. N., “ Reinventing Public Service Delivery : The Case of
environment or system is among the basic characteristics of a Public Information Systems Implementation in Malaysia,” International
good and efficient system and can ensure real time accurate Journal of Public Information Systems, 1, 2009, 9–15.
information [13]. Information system integration is also very [3] Poon, P. and Wagner, C., “Critical success factors revisited: Success
important for the success of e-Government [14] For example, and failure cases of information systems for senior executives,” Decision
integration of several systems in different agencies will enable Support Systems, 30(4), 2001, 393–418. doi:10.1016/S0167-
business license applications to be processed and approved in 9236(00)00069-5
a short time. [4] Elkadi, H., “Success and failure factors for e-government projects: A
case from Egypt,” Egyptian Informatics Journal, 14(2), 2013, 165–173.
Lastly for Construct F (Security), both items F1 and F2 are doi:10.1016/j.eij.2013.06.002
also in the most agreeable category. This shows that the [5] El-far, I. K. and Whittaker, J. A., “Model-based Software Testing,” 1–
security aspect is very important in an information system. 22, 2001.
Nowadays, the introduction and application of Internet of [6] Azrilah Abdul Aziz, Mohd Saidfudin Masodi, and Azami Zaharim,
Things (IoT) in intelligent devices enables the connectivity of “Asas Model Pengukuran Rasch,” Penerbit Universiti Kebangsaan
Malaysia. 2013.
the devices to the Internet [15]. But this has also exposed the
[7] Bond, T. and Fox, C. M., “Applying the Rasch Model Fundamental
devices to security risks such as information leakages or theft. Measurement in the Human Science,” hlm.Second Edi. 2007.
So, it is very important to ensure an information system is [8] Wright, B. D. and Mok, M. M. C., “An Overview of the Family of
secure and prevent system or process failure. Rasch Measurement Models. Introduction to Rasch Measurement,”
2004, 1–24.
VI. CONCLUSION [9] Nopiah, Z. M., Rosli, S., Baharin, M. N., Othman, H., and Ismail, A.,
Overall, this study has successfully identified the factors “Evaluation of pre-assessment method on improving student’s
performance in complex analysis course,” Asian Social Science, 8(16),
and criteria influencing the software quality of information 2012, 134–139. doi:10.5539/ass.v8n16p134
system. By identifying these criteria, it will serve as a
[10] Masum, A.-K., Beh, L.-S., Azad, A.-K., and Hoque, K., “Intelligent
guideline to improve the software aspect of an information human resource information system (i-HRIS): A holistic decision
system and ensure that the system has the capability and support framework for HR excellence,” International Arab Journal of
quality at its best. Information Technology, 15(1), 2018, 121–130.
[11] Anila, M., Sirisha, N., and Karthik, R., “Software reliability models - A
However, technological advancements today has brought comparative study. Proceedings of the International Conference on
about the needs for intelligent information systems that are Intelligent Sustainable Systems,” ICISS 2017, (Iciss), 2018, 1152–1154.
able to perform more complex functions. Thus, there is a need [12] Alshamari, M., “Usability Factors Assessment in Health Information
for future research to look into other software quality factors System,” Intelligent Information Management, 08(06), 2016, 170–180.
and criteria that will give advanced capabilities to the system. [13] Wiemann, S., Brauner, J., Karrasch, P., Henzen, D., and Bernard, L.,
“Design and prototype of an interoperable online air quality information
System developers also play an important role in ensuring system,” Environmental Modelling and Software, 79, 2016, 354–366.
that an information system is able to perform as required. [14] Siti Istianah Mahdzur; and Juhana Salim, “Information Systems
Their inpu is equally important so as to enable the software Integration Factors In Organization: Towards Government Information
quality to be assessed comprehensively. Therefore, it is also Systems Sustainability,” Journal of Theoretical and Applied Information
recommended for future research to take into account the Technology, 71(2), 2015, 235–250.
perspective of system developers. [15] Kumar, S. A.; Vealey, T.; and Srivastava, H., “Security in internet of
things: Challenges, solutions and future directions,” Proceedings of the
Information systems are being used in various fields to Annual Hawaii International Conference on System Sciences, 2016–
assist in decision making and strategic planning. Each field March (January), 2016, 5772–5781.
75 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Finding maximum independent set (MIS) in a Finding an MIS in a graph is considered very useful
graph is considered one of the fundamental problems in the approach for many real life applications and problems, such as
computer science field, where it can be used to provide solutions optimization problems, job scheduling, and social network
for various real life applications. For example, it can be used to analysis. MIS can be determined using brute force approach in
provide solutions in scheduling and prioritization problems. O(N2×2N) run time units, where N is the number of vertices in
Unfortunately, this problem is one of the NP-problems of a graph. This approach takes a lot of time to find an MIS for
computer science, which limit its usage in providing solution for large N as described in [5]. Many approaches and algorithms
such problems with large sizes. This leads the scientists to find a are proposed to find out an MIS of a graph, but with very long
way to provide solutions of such problems using fast algorithms
run time. So, many algorithms have been proposed to find out
to provide some near optimal solutions. One of the techniques
used to provide solutions is to use metaheuristic algorithms. In
an approximation to actual exact MIS solution with less time
this paper, a metaheuristic algorithm based on Chemical complexity as in [6,7,8,2,9].
Reaction Optimization (CRO) is applied with various techniques Here are some definitions related to MIS and CRO:
to find MIS for application represented by a graph. The
suggested CRO algorithm achieves accuracy percentages that An undirected graph is G(V, E), where V is a set of
reach 100% in some cases. This variation depends on the overall vertices and E is a set of edges in G. The set of vertices
structure of the graph along with the picked parameters and is a collection or group that contains the vertices
colliding molecule selection criteria during the reaction (nodes) in the graph and these vertices (nodes) are
operations of the CRO algorithm. connected to each other by links that are called Edges.
The collection or group that contains all graph edges is
Keywords—Chemical reaction optimization; graph; maximum called Edge Set noted by E.
independent set; metaheuristic algorithm; modified Wilf algorithm;
optimization problems An Independent Set (IS) in a graph G(V,E) is defined
in [10,11] to be a set V’, where V' ⊆ V and there is not
I. INTRODUCTION exist an edge that connects vs and ve, where e ∈ E, vs∈
In this paper, a metaheuristic Chemical Reaction V' and ve∈ V' (i.e. either vs or ve∈ V');Where a
Optimization (CRO) algorithm has been utilized to find out Maximum Independent Set (MIS) is defined to be the
maximum independent set (MIS) in a graph. In this approach, IS of the largest size among all available ISs in G.
computational steps are formulated as a set of molecules
Chemical Reaction Optimization (CRO) is defined in
reactions that leads toward approximated solution. CRO
[1,15] as a metaheuristic approach that mimics the
approach considers two types of collisions that could happen:
process of chemical reactions in the field of Computer
On-Wall collision and Inter-molecular collision as illustrated
Science. It relays on minimizing the potential energy to
in [1]. These collisions could be effective or ineffective
depending on the nature and the type of the problem to be the minimal value without sticking in local minima.
implemented or solved. The effective On-Wall collision is This algorithm defines an objective function that is
called decomposition, where the colliding molecule is used to calculate potential energy of the current state of
supposed to be decomposed into several parts (mainly two reaction (execution) process. Just like genetic
parts). Effective inter-molecular collision is called synthesis, algorithms, this is done by iterating for a predefined
which involves merging the colliding molecules together. number of iterations or meeting optimal objective
value.
In [2], Independent Sets (ISs) are described to be some of
the useful information that can be concluded from graphs and On-wall ineffective collision is a CRO operation that
used in real life applications; such as project scheduling and involves colliding the molecule on the wall without any
social network analysis, while they are important concept in effective restructure of the colliding molecule.
building bipartite graphs [3,4] which are fundamental in many Decomposition (On-wall effective collision), is a CRO
computing areas; such as coding theory and projective operation that involves colliding the molecule on the
geometry. Independent set can be defined as a set of nodes in wall effectively so that colliding molecule is
a graph that are not connected. Note that a graph may contain decomposed (divided) into multiple molecules.
several independent sets, and finding the maximum one is the
best goal to achieve. The IS with maximum size is referred to Inter-molecular ineffective collision is a CRO
as MIS. operation that involves colliding two molecules
76 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
together ineffectively so that no major structural O(1.2114n ) to find an exact solution. Such a reduction in the
change would occur. run time could have high influence in case of critical run time
applications like process scheduling on a CPU.
Synthesis (Inter-molecular effective collision) is a CRO
operation that involves colliding two molecules The second approach is using approximation algorithms
together effectively so that a new molecule of the based on heuristics to provide approximate solution in
merged collided molecules will be generated. polynomial-time. According to [12], “Most polynomial-space
algorithms for MIS use the following simple idea to search a
The solution for a problem based on CRO is represented as solution: branch on a vertex of maximum degree by either
a molecular structure noted as (ω), which has a minimum excluding it from the solution set, or including it to the
potential energy that is determined by a problem specific solution set. In the first branch, we will delete the vertex from
objective function noted as PEω, which is determined by an the graph and in the second branch we will delete the vertex
objective function f(ω). Each molecule has a kinetic energy together with all its neighbors from the graph”. Algorithms
that illustrates the tolerance of having worse solutions and that use such heuristics can be found in [7,16]. More
noted as KE. evolutionary heuristic approaches can be found. For example,
In this paper, different techniques are implemented over in [6,17,18,19], genetic algorithm was used to find an
CRO algorithm to provide an approximate solution for the approximate solution for the MIS problem. In [20], a swarm
MIS problem. In these techniques, an implementation of CRO intelligence approach based on ant-colony optimization was
is provided to solve MIS problem to provide near optimal used to find a solution. Note that, approximation algorithms
solution. are used in real applications just like in [21], where genetic
algorithm is used to generate data for testing PLSQL
In the remaining sections, a review of related work is (Procedural Language extension to Structured Query
presented in Section 2. A description of the proposed Language) program units. This generated data is a sub-set of
algorithms will be explained in Section 3. This is followed by the actual data range that can’t be covered in some extreme
experimental results in Section 4 and discussion in Section 5. cases, where data to be tested is huge and can’t be tested using
Section 6 presents the conclusion and intended future normal brute-force concept. A more generic test data
research. generation for software testing is proposed in [22] to generate
In this paper, a new approach is applied to find Maximum test data using genetic algorithm for software testing purposes
Independent Set to explore its ability in order to find better rather than using normal brute-force test data generation.
approximation results than previous approaches that cannot be The third approach is using exact algorithms to find exact
applied on huge graphs which may contain millions of nodes. solution in polynomial-time, but for graphs of special classes,
Finding a maximum independent set with near optimal results such as designing a polynomial run time algorithm that finds
would be used to provide a solution of many real-life an exact solution in graphs with vertex of degree 2 at
applications; such as prioritization and scheduling maximum. Such algorithms are case sensitive ones and can’t
applications. be generalized to find exact solutions to graph of random
shape and arbitrary degree. Examples of this form of
II. RELATED WORK
algorithms can be found in [23], where an exact algorithm is
In their research for finding solution of the MIS problem, provided for graphs with vertices of maximum degree of 3, or
researchers have handled the issue using different approaches in [12,24], where, in addition to an exact solution provided for
based on type of final result or the nature of graph, such as any random graphs, the authors provided a O(1.1571 N ),
degree of nodes, as illustrated in [10,11,12] where such 1.1737N×NO(1), 1.1893N×NO(1), and 1.1970N×NO(1), for graphs
approaches have been used. In general, finding MIS can be of maximum degree of 4, 5, 6, and 7, respectively.
done using one of the following three approaches: using brute
force algorithm, approximation algorithms, and exact As mentioned before, all exact solutions attempts consume
algorithms for special type of graphs. a very large amount of time to execute. Such algorithms
would decrease the feasibility of e solutions. So, a new
The first approach is using an exact (brute force) paradigm of computing near optimal solution has been
algorithm. The direct way to solve such problem is to check proposed, such as in [7,8,9,16]. As illustrated in [8], this is
all possible solutions by representing the presence of node in done using heuristic or metaheuristic techniques. Combining
the solution by 1 and the absence by 0 as mentioned in [11]. of exact and meta-heuristic algorithms can provide near
So, we can represent the solution by a binary number with optimal solution in a shorter time like in [25] where better
length N, where N is the number of nodes in the given graph. execution time has been achieved. Moreover, there are some
This involves checking 2N numbers that represent all possible known strategies to do parallel implementation of
subsets of the original set of nodes. For each solution (binary metaheuristic approaches. By parallelizing these algorithms
number), all nodes must be checked to ensure disconnection of such as in [26], an enhanced version with better performance
nodes (N2). So, final run time complexity would be O(N2×2N). could be achieved.
Nevertheless, some researchers have produced exact
algorithms with better runtime. In [14], the authors proposed In [27], CRO has been used to find optimal solution for
an algorithm, which achieved an exact solution in O(1.2132 n) task scheduling and resource allocation in grid computing.
time for a graph of size n vertices, while in [15], the authors They propose several versions of CRO to solve task
provided an algorithm with running time complexity of scheduling problem. These versions have been experimented
77 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
and tested against four metaheuristic approaches. The results In [36], CRO has been used to solve Set Covering Problem
show that CRO outperforms the other approaches in terms of (SCP) while in [37] a strengthened version of clique covering
accuracy and performance, especially in case of large test has been investigated. SCP can be formulated as the
instances. following:
In [1], CRO has been used to provide a solution for Given a set M, ⊆ , j = 1,...,n are n subsets of M,
quadratic assignment problem described in [28]. CRO and weights of the subsets, cj, j = 1,...,n; and set cover
implementation has been tested against various evolutionary is a collection ⊆ {1,...,n} such that ⋃ ∈ = M.
approaches. Test results show that CRO implementation SCP tries to minimize the cost of covering the entire
outperforms other implementations in many cases. Parallel set using a subset of the original set. There are two
implementation of CRO has been used to solve the same types of set covering problem, unicost, and non-
problem in [29], where test results show that parallel CRO unicost. CRO outperformed the accuracy of other
implementation provides better performance along with algorithms in case of non-unicost SCP, where optimal
solution quality in comparison with sequential one. solution has been determined in 65 experiments. In
In [1], CRO algorithm has been used to solve resource- case of unicost SCP, CRO shows outstanding
constrained project scheduling problem described in [30] as performance in comparison to other approaches.
planning the project milestones according to predefined In [38], a version of CRO called Greedy CRO (CROG) has
priorities. In real life, project is divided into fixed time slots. been proposed and implemented to solve 0-1 Knapsack
Project activities are assigned to time slots according to Problem. Experimental results show that CROG outperforms
available resources that are limited, while activities could be other metaheuristic approaches, such as genetic algorithms,
dependent on each other. CRO is used to find best scheduling ant-colony, and quantum-inspired evolutionary algorithms.
of tasks that minimizes project lifetime. Test results show that
CRO implementation can achieve better results for known In [39], enhanced version of CRO has been used to find
benchmarks. optimal road network design that takes into consideration the
cost along with noise and vehicles emissions. Proposed CRO
In [1], CRO has been used to provide a solution for was tested against Genetic Algorithm (GA) for comparison.
channel assignment problem in wireless mesh networks Test results show that CRO outperformed GA in most cases.
described in [31] to assign available channels to multiple
wireless networks. It is used for wireless communication In [40], Objective Power Flow (OPF) problem has been
channel selection to be used in the communication between solved using CRO algorithm. OPF aims to minimize power
neighboring mesh routers without suffering any interference generation cost by considering many constraints, such as the
or communication problems. The results show that CRO has balance of the power, bus voltage magnitude limits,
improved current solutions of the problem. transmission line flow limits, and transformer tap settings. The
results show that CRO can provide the best results among
In [32], CRO has been used to solve population transition other algorithms on the IEEE-30 test case. Note that best
problem in peer-to-peer live streaming. In this problem, result is the one with lowest power flow cost.
network live streaming has been improved by grouping peers
into multiple colonies according to delay. Peers with less In [27], CRO implementation has been extended using
delay can act as service providers for longer delay ones. So, parallel approach to solve the Quadratic Assignment Problem
the system is said to be in universal streaming when all peers (QAP). QAP seeks to optimally assign facilities to locations
are served with sufficient streaming data. Test results show in a way to minimize transportation cost of facilities, as they
that evolutionary approach of CRO outperforms existing non- are required in multiple locations. Parallel CRO has been
evolutionary approaches. compared with sequential one in solving QAP, experimental
results show that parallel CRO reduces computational time
In [33], CRO has been used to find a solution for network with more accurate results.
coding optimization problem described in [34] to provide
coding mechanism for network with minimum number of In [41,42], CRO implementation has been done to solve
digits. In this problem, network coding has been used to Max Flow problem (MFP) in a way that is close to Ford-
enhance transmission rate between routers on certain Fulkerson algorithm. In [42], the results have been compared
interfaces. This strategy of coding specific interfaces could with GA in term of accuracy and performance. The results
increase transmission rate without avoidance of extra show that the problem is solvable by CRO and GA; however,
computational overhead by coding all available interfaces. the GA one outperforms the CRO one.
Test results show that CRO outperformed existing algorithms. In this research, we provide adapted versions of CRO to
In [35], CRO has been used in Artificial Neural Network find a solution of the MIS problem. Several scenarios are
(ANN) training. ANN is composed of layers that contain investigated when a molecule (subset of the graph) is selected
multiple computational units called neurons. Neurons must be randomly among available molecules, and a molecule is
assigned weights to provide best results. Tuning is done by selected according to certain criteria. The selected criteria are
training the network with set of training data. Test results the minimum connectivity. The adapted CRO algorithm with
show that CRO trained ANN has better testing error. its implementation and performance are presented.
78 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
III. CRO ALGORITHM FOR MIS that collided molecule should be decomposed into two
In CRO, a molecule is represented by a node in a graph. molecules. Regardless of collision type or its effectiveness, all
Thus, an MIS has set of not connected nodes, or set of none affected molecules potential energy should be calculated and
neighboring nodes. In such representation, the CRO considers checked with previously registered minimum value of the
each molecule a candidate solution (i.e. Independent Set). molecules.
Molecule potential energy is defined as the number of In case of on-wall collision, effectiveness of the collision
remaining graph nodes that are not contained in the molecule. depends on how many times a molecule collision did happen
So, if the number of graph nodes is 50 and the molecule without any improvement in the solution. So, if a predefined
contains 5 nodes, the potential energy is 50-5=45. Fig. 1 number of iterations are reached without any improvement in
shows the flowchart of the CRO algorithm. its minimum value, the collision is defined to be an effective
Initially, there are N molecules manipulated by the on-wall collision. In this case, the molecule is divided into two
algorithm, since every node is considered as one molecule, molecules, where each molecule contains the same number of
which is the minimum solution (each node is an independent original molecule’s nodes. For example, if the collided
set). A molecule is selected for the purpose of collision in each molecule contains nodes {1, 10, 19, 50}, this molecule will be
iteration. Collision type is selected according to the initial divided into two molecules one molecule contains {1, 10},
inter-molecular to on-wall collisions ratio. while the other one contains {19, 50}.
In case of inter-molecular collision, effectiveness of the Another main factor of the proposed algorithm is molecule
collision depends on whether selected molecules can be selection, which indicates to how a molecule is selected for
merged together or not. This is done by checking the further processing, such as on-wall collision or inter-molecular
confliction between the two molecules, so that each node in collision. In this proposed algorithm, multiple scenarios are
the second molecule is checked with the conflicting (i.e. tested, as the following:
neighbors) nodes of the first molecule. If the node is found 1) A molecule is selected randomly among available
among the conflicting nodes of the first molecule, the collision molecules.
is defined to be none effective collision and nothing would 2) A molecule is selected according to certain criteria.
happen because the two molecules are not eligible to be
The selected criteria are the minimum connectivity.
merged. This is because each molecule is assumed to be an
independent set, and it is not allowed to contain conflicting TABLE. I. MAPPING CHEMICAL REACTION TO MATHEMATICAL
nodes. On the other hand, if the entire nodes of the second MEANING
molecule are not exist among the conflicting nodes of the first
molecule, the collision is defined to be effective, so that Chemical Mathematical
Mathematical Meaning
selected molecules are merged together and new molecule is Meaning Representation
formulated. This new molecule contains the whole nodes of
the collided molecules. Molecular
Solution Ω (e.g. MIS)
structure
Table I shows the mapping of chemical notations to their
corresponding mathematical representation defined in [1,13].
The solution is represented by a molecular structure noted as PEω = f(ω) (e.g. number of
remaining nodes in a graph
(ω). Potential energy Objective function value
that are not selected as in
the solution)
Fig. 1 was adapted from [1,13], shows a flowchart of the
CRO algorithm, which indicates that the first step of the
algorithm is the initialization, as described in [1]. Initialization Measure of tolerance of
KEω(e.g. the same value
includes pre-processing (e.g. preparing the data in appropriate Kinetic energy determined by the original
having worse solutions
data structure, and removing unnecessary data), and initial algorithm)
values calculations (e.g. algorithm variables and constants).
This step is followed by the iteration checking condition, Number of hits
Current total number of
(e.g. number of iterations)
which examines stopping criteria condition to avoid infinite moves
calculations or iterations. If stopping criteria condition is met,
the algorithm execution is finished, and no more iteration is Minimum
(e.g. the best solution found
done. On the other hand, if the condition is not satisfied, no Current optimal solution during the execution of the
structure
algorithm)
more iteration is done. In each iteration, a collision must be
performed, which could be either on-wall or inter-molecular
collision. This involves determination of which action to be Current optimal function (e.g. the potential energy of
Minimum value
value the minimum structure)
taken in the next iteration. If the collision type is selected, the
next step is to decide whether the selected collision type is
effective or ineffective according to the selected collision Minimum hit
Number of moves when (e.g. number of iterations
molecules. In case of intermolecular collision, effective the current optimal “hits” till finding the
number
solution is found minimum structure)
collision is called synthesis, which indicates that collided
molecules should be merged. In case of on-wall collision,
effective collision is called decomposition, which indicates
79 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
According to the above criteria of molecule selection, iteration. In case of inter-molecular collision iteration, the
multiple combinations are tested to find out whether things go second molecule is selected randomly.
better or not, as follows: 5) Single minimum molecule with minimum degree
1) Multiple random molecules: In this scenario, molecules molecules: In this scenario, the same behavior of the previous
are selected randomly for collision. In each iteration, a random scenario (4) is done with a difference that the second molecule
molecule is selected for collision with another random in case of inter-molecular collision is selected according to its
selected molecule, or to collide with wall. connectivity degree, so that minimum connectivity degree is
2) Single random molecule with random molecules: In selected to collide with initial minimum connectivity degree
this scenario, a random molecule is selected as a main molecule.
molecule. In each iteration, this molecule is selected as the If the collision is defined to be an effective inter-molecular
main molecule. In case of inter-molecular collision is collision, the components of the molecule are merged together
performed, the second molecule is selected randomly. So, in and the conflicting nodes are computed with redundant nodes
this scenario, all iterations are done on the same molecule, but removal (no redundancy in conflicting nodes). The old
the variation appears in the second molecule only. molecules are removed from the pool of available molecules,
3) Single random molecule with minimum degree while the resultant molecule is added to the pool.
molecules: This scenario appears to be the same as the In an iteration, potential energy is updated according to
previous one, where a single random starting molecule is equation (1).
selected and used for every iteration in the reaction life cycle.
(1)
But, the variation is that second molecule in case of inter-
molecular collision is selected according to the criteria that is Where ω denotes a molecule, Size(ω) denotes number of
not random. Instead of that, the second molecule is selected nodes in a molecule, and N denotes the number of nodes in the
according to its connectivity degree, where minimum graph.
connectivity degree molecule is selected to collide with fixed Kinetic energy doesn’t affect the process of CRO in this
starting random molecule. proposed algorithm, since each molecule is assumed to be
4) Single minimum molecule with random degree effective and capable of reacting with other molecules at any
molecules: In this scenario, minimum connectivity degree moment, regardless of its situation or kinetic energy.
molecule is selected at the beginning and used for every
80 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
81 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Name: chooseMinimumConflectingMolecule molecule. So, initially, there are 5 molecules where these
Input: Molecule object, and available molecules list. molecules contain nodes 1, 2, 3, 4, and 5; while potential
Output: Chosen Molecule object (Minimum Connectivity Degree Molecule). energy for each of them is 4. This is because there is a graph
Function chooseMinimumConflectingMolecule(Molecule molecule){ node in the molecule, and the remaining graph nodes are not
Molecule pickedMolecule=null
if(molecule == null){
included in the molecules.
for each molecule in the available molecules { The algorithm will pick molecule with minimum
if(the molecule contains the default selected node){
pickedMolecule = current molecule
conflicting nodes first, and do all reactions on that molecule.
Break In this example, the algorithm can pick one of the molecules
} containing nodes 1, 4, and 5, as each has minimum number of
} conflicting nodes, which equals to 2.
}
else { Name: initmc
Molecule minimum = null Input: Graph Nodes.
int min = Number of nodes in graph Output: initializing molecules (conversion of graph nodes into CRO
for each molecule in the available molecules { molecules) and picks minimum connected node molecule as initial starting
if (currentMolecule != molecule){ solution.
int temp = number of conflecting nodes in currentMolecule function initmc(){
if ((minimum == null) or (temp < min)){ noOfIterations = number of graph nodes
minimum = choosenMolecule //the algorithm will iterate exactly the number of nodes
min = temp minimumNoOfIterations = 0
} //initial minimum number of iterations to find solution is 0
} minimumSize = number of graph nodes
} /*minimum solution initially is same number of graph nodes (maximum
pickedMolecule = minimum excluded nodes in worst case)*/
} int minimumLinks = number of graph nodes + 1
return pickedMolecule /*initial minimum number of node linkes is the number of graph nodes +1
} note that this variable is used to keep track of discovered minimum no of
node neighbors*/
Fig. 6. Chooseminimumconflectingmolecule Function that Chooses foreach node in the graph nodes{
Minimum Connectivity Degree Molecule from Available Molecules. MISMolecule molecule = create molecule containing current graph node
only
Name: initr //each node in the graph would be represented as a unique molecule.
Input:Graph Nodes molecule.PotentialEnergy = number of graph nodes-1
Output: initializing molecules (conversion of graph nodes into CRO /*initial molecule potential energy is the no of remaining graph nodes
molecules). not included in the molecule which is the number of graph nodes-1*/
Function initr(){ molecule.NumberOfHits = 0
noOfIterations = number of graph nodes //initial no of hits is 0 where no collisions have occurred.
minimumNoOfIterations = 0 molecule.MinimumHitNumber = 0
minimumSize = number of graph nodes //minimum no of hits to find best solution is initially 0
intselectedIndex= pick random number between 0 and number of nodes-1 molecule.MinimumStructure = molecule;
foreach node in the graph nodes{ /*minimum structure (best solution) is the initial one which is the current
MISMolecule molecule = create molecule containing current graph node molecule structure (one node)*/
only molecule.MinimumValue = molecule.PotentialEnergy
molecule.PotentialEnergy = number of graph nodes-1 /* minimum value of potential energy (best solution value) is the initial
molecule.NumberOfHits = 0 one which is the initial potential energy of molecule*/
molecule.MinimumHitNumber = 0 add molecule to the available molecules
molecule.MinimumStructure = molecule; //adding molecule to the molecules pool.
molecule.MinimumValue = molecule.PotentialEnergy if (minimumLinks> number of node Neighbors)
add molecule to the available molecules {
if (node index=selectedIndex){ selectedNode = node
selectedNode = node minimumLinks = number of node Neighbors
} }
} /*check the number of current node neighbors so that if it is less than
remove molecules that contain selected node neighbors nodes from the minimum observed links, then its corresponding molecule will be
available molecules selected to be initial colliding molecule and its number of neighbors is
noOfIterations = noOfIterations - number of removed molecules saved in minimumLinks to keep track of it and compared to remaining
} nodes*/
}
Fig. 7. Initr Function that Initializes the Execution of CRO Algorithm and remove molecules that contain selected node neighbors nodes from the
Chooses Starting Molecule Randomly. available molecules
/*selected node neighbors should be excluded from the molecules pool
B. Example since they won’t be part of the solution (IS) since their neighbor node is
selected to be initial part of the solution*/
In this section, an example of the algorithm execution is noOfIterations = noOfIterations - number of removed molecules
provided by considering (Minimum initial node &Minimum //number of iterations decreased by the number of removed molecules
iteration node) algorithm. Consider the graph in Fig. 9. The }
algorithm will initialize CRO molecules by representing each Fig. 8. Initmc Function that Initializes the Execution of CRO Algorithm and
graph node by a single molecule. The potential energy equals Chooses Starting Molecule with Minimum Connectivity Degree.
to the number of remaining nodes not included in the
82 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
83 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
(3)
Complexity of collision effectiveness computation:
The overall run time complexity of the collision is the
1) On-wall collision: As described in [1,13,32], the complexity of collision effectiveness calculation, the molecule
effectiveness of on-wall collision is determined by checking selection complexity, and the collision execution complexity
the number of ineffective iterations of the molecule. according to its effectiveness; and is expressed as in equation
Ineffective iterations are the iterations that have been done on (4).
the molecule after minimum value is found without any
improvement. If the number of iterations exceeds a predefined
(4)
constant value, the collision will be defined to be an effective
one; otherwise it is not. So, the run time complexity of In case of On-Wall collision, there are two cases:
determining effectiveness of the collision is constant.
1) Ineffective collision: By applying equation (4), the
2) Inter-molecular collision: The effectiveness of the
resultant is equation (5) for On-Wall collision complexity
collision is determined by checking the readiness of molecules
to be merged together. This is done by checking the existence (5)
of any of the second molecule nodes within the first molecule Where O(Mol. Selection) depends on the molecule
conflicting nodes. If the check determines that any of the selection criteria. So, in case of random molecule selection,
second molecule nodes exists in the conflicting nodes of the the resultant equation is represented in equation (6). While in
first molecule, the collision is defined to be effective; case of minimum connectivity degree molecule selection is
otherwise, it is not. In the worst case, first molecule contains used, the collision run time complexity is as in equation (7).
one node and (N-1) conflicting nodes, while the second
(6)
molecule contains all the remaining graph nodes so that its
size is (N-1). To check the existence of second molecule nodes (7)
in the conflicting nodes of the first molecule, the whole list of
2) Effective collision: By applying equation (4), the
the first molecule nodes should be iterated for every node in
resultant equation (8) of collision complexity is as in equation
the second molecule, until finding the checked node or
(7).
reaching the end of the list and the node is assumed to be not
conflicting. So, the iterations are, [1,2,3,…,N-1] and the run (8)
time complexity is [1+2+3+…+(N-1)].In this case, the run Where O(Mol. Selection) depends on the molecule
time complexity of the effectiveness calculation is O((N-2)(N- selection criteria. So, in case of random molecule selection,
3)/2)=O(N2). the resultant equation is as in equation (9). While in case of
Complexity of molecule selection types: minimum connectivity degree molecule selection is used, the
collision complexity is as in equation (10).
1) Random selection: In the random selection, the
(9)
algorithm will pick a random molecule from the list of
available molecules to perform intended operation. So, in this
case, no processing is done, and a constant number of steps (10)
(K) is performed.
In case of Inter-Molecular collision, there are two cases:
2) Minimum connectivity degree node selection: In this
case, the algorithm will iterate through the available molecules 1) Ineffective collision: By applying equation (4), the
to select the molecule with minimum number of conflicting resultant collision complexity is as in equation (11). The
nodes. In the worst case, the number of molecules is equal to ineffective collision does not perform any operation on the
the number of graph nodes (N). So, the algorithm will iterate colliding molecule(s). So, the complexity of its execution is
through N molecules to find out the one with minimum constant (K). But the calculation of collision effectiveness in
number of conflicting nodes. The complexity of finding worst-case would check the half of graph nodes against the
minimum connectivity degree among N nodes is O(N). The second half of graph nodes that could be fully connected. So,
run time complexity of finding the same initial molecule is the final equation would look like the following:
O(N). ∑
One of the main constants to be defined prior to algorithm (11)
execution is Inter-Molecular to On-Wall collisions Ratio (R).
According to the value of R, the number of inter-molecular Where O(Mol. Selection) depends on the molecule
collisions equals to R (Number of CRO iterations) and on- selection criteria. So, in case of random molecule selection,
wall collisions will be (1-R) (Number of CRO iterations). the resultant is as in equation (12). While in case of minimum
So, equations (2) and (3) will hold. connectivity degree molecule selection is used, the collision
complexity is as in equation (13).
(2)
(12)
84 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
85 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. EXPERIMENTAL EXECUTION TIME IN MSEC. FOR 20%, 60 %, 80%, AND 90% CONNECTIVITY DEGREES’ GRAPHS WITH INITIAL RANDOM
GRAPH NODE SELECTION AND RANDOM MOLECULE SELECTION (RR) IN EACH ITERATION WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
Connectivity
20% 60% 80% 90%
Degree
Size/Collision
0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95
Ratio
100 593 362 256 222 50 59 46 50 19 16 19 12 6 6 3 9
200 4619 3209 1912 1925 272 256 246 271 28 53 53 49 15 15 15 16
300 13675 9078 5958 6147 900 698 831 872 137 128 121 156 37 31 34 28
400 35356 22709 15665 14917 1791 1493 1725 2094 212 234 262 281 44 47 47 46
500 62235 47346 29810 30129 4531 2631 3256 4047 218 412 409 528 84 75 84 87
600 117545 72077 48263 51952 8056 4695 5490 6882 968 625 825 912 75 112 118 134
700 187109 126150 79830 82306 10953 8171 8324 11051 1250 953 1122 1424 140 140 165 193
800 269999 192701 107425 121437 12533 11324 12495 16707 831 1297 1787 2106 237 187 240 281
900 407757 277431 166954 177255 18281 14379 18175 22828 2515 1796 2368 2940 124 234 356 406
1000 551154 373748 232953 242983 26922 20995 24464 32371 2762 1965 3231 4069 281 381 431 544
TABLE. III. EXPERIMENTAL EXECUTION TIME IN MSEC. FOR 20%, 60 %, 80%, AND 90% CONNECTIVITY DEGREES’ GRAPHS WITH INITIAL MINIMUM
GRAPH NODE SELECTION AND RANDOM MOLECULE SELECTION (MR) IN EACH ITERATION WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
Connectivity
20% 60% 80% 90%
Degree
Size/Collision
0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95
Ratio
100 606 409 287 251 109 81 65 69 28 25 21 18 19 12 6 10
200 4230 3035 2130 2057 553 379 296 367 90 78 95 91 43 28 34 26
300 14841 11393 7000 6832 1234 906 991 1198 212 250 253 280 59 53 59 57
400 32222 24763 14987 15814 3309 2185 2100 2555 496 303 381 451 77 87 93 106
500 67398 45029 31119 30814 4398 3722 4239 5010 543 510 793 925 168 122 178 183
600 121295 75942 49422 54483 6157 5507 7376 9413 700 862 925 1171 122 206 224 246
700 189972 137178 87550 89818 12060 10264 11134 13771 1034 1090 1503 1940 412 275 331 425
800 266150 188499 123894 130012 24717 15082 16855 21770 1847 2119 2647 3414 390 365 478 597
900 409454 267807 161301 187702 24989 19514 23766 31255 4853 2406 3150 3885 432 453 621 757
1000 545105 389631 229173 256270 35639 26005 27629 37977 2232 3428 4365 5817 409 568 797 1044
TABLE. IV. EXPERIMENTAL EXECUTION TIME IN MSEC. FOR 20%, 60 %, 80%, AND 90% CONNECTIVITY DEGREES’ GRAPHS WITH INITIAL RANDOM
GRAPH NODE SELECTION AND MINIMUM MOLECULE SELECTION (RM) IN EACH ITERATION WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
Connectivity
20% 60% 80% 90%
Degree
Size/Collision
0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95
Ratio
100 476 328 212 193 80 47 43 46 21 15 12 9 7 6 3 0
200 3917 2423 1525 1439 299 218 241 218 56 34 34 31 21 21 15 12
300 5952 5385 4818 4869 753 616 679 650 134 87 100 93 34 31 31 28
400 23999 16660 11689 11410 975 1113 1263 1486 362 228 196 203 56 44 34 37
500 52396 34185 24442 22005 3730 2803 2370 2844 478 359 325 387 62 53 56 68
600 34862 36307 38833 38067 4160 3831 4451 4900 375 353 515 669 68 81 106 97
700 94925 101125 54167 61317 5401 6825 7222 7741 356 750 903 968 81 97 128 149
800 233388 173169 94075 91432 2106 8976 10587 11448 1053 697 1262 1478 200 206 200 231
900 334125 149173 143130 129653 7993 16879 14036 16306 1284 1028 1750 2112 259 253 250 284
1000 309296 230572 168135 176303 8178 10025 19302 22409 1659 2281 2365 2872 331 337 322 365
86 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. V. BEST ACCURACY RESULTS OF INITIAL RANDOM GRAPH NODE SELECTION AND RANDOM MOLECULE SELECTION (RR) IN EACH ITERATION
ON 20%, 60%, 80%, AND 90% CONNECTIVITY DEGREE GRAPHS WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
Fig. 10. Average Execution Time for All Tested CRO Algorithm Versions with (Collision Ratios 0.25, 0.5, 0.75, and 0.95).
In Table VI, we demonstrate the accuracy of CRO accuracy is dropping when the number of nodes is growing
algorithm running using full random selection of initial and up. When the connectivity degree increases, the accuracy
iteration molecules, where the accuracy is calculated using becomes more stable and near to constant regardless of graph
equation (20) according to Modified Wilf algorithm results. In size. Moreover, when CRO algorithm is run using 95% inter-
20% connectivity degree, Modified Wilf algorithm can obtain molecules to on-wall collision ratio, it provides better results.
an MIS from a graph of up to 150 nodes in an acceptable time.
While in the higher connectivity degrees (60%, 80%, and Table VIII shows accuracy results of CRO algorithm using
90%) the solutions are obtained in a graph of up to 1000 random selection of initial iteration molecules and picking
nodes. The results show that the accuracy is dropping when minimum connectivity node in each iteration, where the
the number of nodes is going up. When the connectivity accuracy is calculated using equation (20) according to
degree increases, the accuracy becomes more stable and near Modified Wilf algorithm results. In the higher connectivity
degrees (60%, 80%, and 90%), the solutions are obtained from
to constant regardless of graph size. Moreover, when CRO
algorithm is run using 95% inter-molecules to on-wall graph of up to 1000 nodes. The results show that the accuracy
collision ratio, it provides better results. This is a result of is dropping when the number of nodes is growing up. This is a
performing more inter-molecule collisions, which provides normal result of increasing the number of nodes, where the
more combinations of nodes (solutions), so that better size of MIS becomes greater, so that the percentage won’t be
solutions could be discovered. affected by low number of nodes, not like small solutions,
where a single node could increase the percentage of accuracy
Table VII demonstrates the accuracy of CRO algorithm by a significant value.
using random selection of iteration molecules, while starting
with minimum molecule (minimum connected node), where Table IX shows the accuracy results of CRO algorithm
the accuracy is calculated using equation (20) according to using minimum connectivity molecule and selecting minimum
Modified Wilf algorithm results. The results show that the connectivity molecule in each iteration, where the accuracy is
calculated using equation (20) according to Modified Wilf
87 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
algorithm results. In the higher connectivity degrees (60%, Extra experiments have been done to test proposed
80%, and 90%), the solutions are obtained up to 1000 nodes implementation on benchmark datasets, such as Graph50_10,
graph size. The results show that the accuracy is dropping Graph100_10, Hamming6_2, Hamming6_4, Hamming8_4,
when the number of nodes is going up. Moreover, the and Hamming10_4 obtained from [43,44,45]. The results
algorithm shows almost identical accuracy regardless of inter- listed in Table X show that the CRO algorithm provides
molecule to on-wall collisions ratio. This indicates that the optimal solution in some cases, specially, when the selection
best results are obtained early at the beginning of execution, of molecules is done in random and the inter-molecular to on-
so that it doesn’t differ if the collisions between the molecules wall collisions ratio is high, such as (75% or 95%). On the
are increased or not. This indication can be used to decrease other hand, the results show that minimum degree molecule
the number of iterations in case of higher ratio; but the selection criteria provide lower accuracy, which tends to be
problem is how to obtain stopping condition? the result of selecting special molecules each time of collision,
which could deviate from the correct path of optimal solution
Fig. 11 shows the average accuracy of each type of that may contain higher degree nodes. The results show that
algorithms along with inter-molecule to on-wall collisions optimal solution (Exact solution) of MIS could be achieved by
ratio. The figure shows that random selection of molecules in CRO. But, the main problem is that this result is not
CRO iterations provides better accuracy results, especially, guaranteed. CRO should be executed many times (in our case
when the ratio of inter-molecule to on-wall collisions 10 times) to have more solutions that may contain the optimal
increases. The algorithm performance on graph with95% ratio one. So, if the execution of CRO is finished within 1 second,
provides better results in case of random selection. These and the re-execution is done 10 times, this means that the total
results represent the worst results in term of accuracy among execution time is 10 seconds, which is the actual time to be
all tests. This is a result of using minimum number of compared with. This makes Modified-Wilf better choice and
neighbors as selection criteria for initial base molecule and more worthy to use in case of small problems (lower graph
other molecules in each iteration, so that static selection of size and higher connectivity), since the difference of achieved
colliding nodes is performed, and less nodes combinations are
performance is low with guaranteed results.
discovered.
TABLE. VI. BEST ACCURACY RESULTS OF INITIAL MINIMUM GRAPH NODE SELECTION AND RANDOM MOLECULE SELECTION (MR) IN EACH ITERATION
ON 20%, 60%, 80%, AND 90% CONNECTIVITY DEGREE GRAPHS WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
TABLE. VII. EXPERIMENTAL EXECUTION TIME IN MSEC. FOR 20%, 60 %, 80%, AND 90% CONNECTIVITY DEGREES’ GRAPHS WITH INITIAL MINIMUM
GRAPH NODE SELECTION AND MINIMUM MOLECULE SELECTION (MM) IN EACH ITERATION WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
Connectivity
20% 60% 80% 90%
Degree
Size/Collision
0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95 0.25 0.5 0.75 0.95
Ratio
100 602 385 266 209 127 84 75 50 27 18 12 12 16 15 15 6
200 1002 1493 1562 1616 622 427 293 262 54 46 62 68 36 31 21 22
300 13926 8450 6857 5362 299 403 719 912 384 275 206 200 53 40 43 43
400 38450 23482 13688 11688 2965 2118 1897 1776 600 440 331 319 87 62 75 81
500 49903 36941 26663 23738 972 1603 2972 3635 1228 837 671 653 215 200 128 131
600 113015 70506 47772 40013 15934 10204 7210 6739 1650 1290 937 802 87 121 159 190
700 90279 80873 63352 66949 2159 4441 7888 9837 362 794 1084 1406 153 187 284 293
800 138417 97888 83193 98044 29062 19616 15420 15520 3347 2821 2600 2434 221 268 322 447
900 384084 229520 160319 139371 3503 9079 16972 22223 772 1409 2290 2816 1187 737 615 525
1000 27344 79701 140458 187492 41688 37861 27133 26490 959 2088 3381 4256 225 400 603 718
88 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. VIII. BEST ACCURACY RESULTS OF INITIAL RANDOM GRAPH NODE SELECTION AND MINIMUM MOLECULE SELECTION (RM) IN EACH ITERATION
ON 20%, 60%, 80%, AND 90% CONNECTIVITY DEGREE GRAPHS WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
TABLE. IX. BEST ACCURACY RESULTS OF INITIAL MINIMUM GRAPH NODE SELECTION AND MINIMUM MOLECULE SELECTION (MM) IN EACH
ITERATION ON 20%, 60%, 80%, AND 90% CONNECTIVITY DEGREE GRAPHS WITH (COLLISION RATIOS 0.25, 0.5, 0.75, AND 0.95)
Fig. 11. Average Accuracy for the Tested CRO Algorithm Versions with (Collision Ratios 0.25, 0.5, 0.75, and 0.95).
89 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. X. SIZES OF MIS RESULTED FROM EXECUTING CRO ALGORITHM ON A SELECTED SET OF BENCHMARK DATASETS
90 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[19] S. Mehrabi, A. Mehrabi, and A. Mehrabi, “A New Hybrid Genetic [34] M. Kim, M. Medard, V. Aggarwal, U. OReilly, W. Kim, and C. Ahn,
Algorithm for Maximum Independent Set Problem”. In Proceedings of “Evolutionary approaches to minimizing network coding resources”.
the 4th International Conference on Software and Data Technologies, Proceedings of the 26th annual IEEE conference on computer
(ICSOFT 2009), Sofia, Bulgaria, July 26-29, 2009, pp: 314 – 317, Communications, Anchorage, AK, USA,2007,
https://doi.org/10.5220/0002253403140317. https://doi.org/10.1109/INFCOM.2007.231.
[20] L. Youmei and X. Zongben, “An Ant Colony Optimization Heuristic for [35] P. Palmes, T. Hayasaka, and S. Usui, “Mutation-based genetic neural
Solving MIS Problems, “Computational Intelligence and Multimedia network”. IEEE Trans Neural Network, 2005, 16(3), pp:587–600,
Applications, ICCIMA 2003. Proceedings. Fifth International https://doi.org/10.1109/TNN.2005.844858.
Conference, pp: 206-211, 2003, http://dx.doi.org/10.1109/ [36] J. Yu, A. Lam, and V. Li, “Chemical reaction optimization for the set
ICCIMA.2003.1238126. covering problem”. in Proceedings of 2014 IEEE Congress on
[21] M. Alshraideh, B. Mahafzah, H. Salman, and I. Salah, “Using genetic Evolutionary Computation (CEC 2014), Beijing, China, 6-11 July 2014,
algorithm as test data generator for stored PL/SQL program units”, In IEEE CEC Proceedings, 2014, pp: 512-519,
Journal of Software Engineering and Applications,2013, 6(2), pp: 65-73, https://doi.org/10.1109/CEC.2014.6900233.
http://dx.doi.org/10.4236/jsea.2013.62011. [37] Y. Shang, "Poisson approximation of induced subgraph counts in an
[22] M. Alshraideh, B. Mahafzah, and S. Al-Sharaeh, “A multiple-population inhomogeneous random intersection graph model". Bulletin of the
genetic algorithm for branch coverage test data generation”, Software Korean Mathematical Society, in press.
Quality Journal, 2011, 19(3), pp: 489-513, [38] T. Truong, K. Li, and Y. Xu, “Chemical reaction optimization with
https://doi.org/10.1007/s11219-010-9117-4. greedy strategy for the 0–1 knapsack problem”. Applied Soft
[23] I. Razgan, “Faster Computation of MIS and Parameterized Vertex Cover Computing, 2013, 13(4), pp: 1774–1780,
for Graphs with Maximum Degree 3,” Journal of Discrete Algorithms, https://doi.org/10.1016/j.asoc.2012.11.048.
2009, 7(2), pp: 191-212, https://doi.org/10.1016/j.jda.2008.09.004. [39] W. Szeto, Y. Wang, and S. Wong, “The chemical reaction optimization
[24] M. Xiao and H. Nagamochi, “An exact algorithm for maximum approach to solving the environmentally sustainable network design
independent set in degree-5 graphs”. Discrete Applied Mathematics 199, problem”. Computer-Aided Civil and Infrastructure Engineering, 2014,
2016, pp: 137–155, https://doi.org/10.1016/j.dam.2014.07.009. 29(2), pp: 140-158, https://doi.org/10.1111/mice.12033.
[25] J. Puchinger and G. Raidl, “Combining Metaheuristics and Exact [40] Y . Sun, A. Lam, V. Li, J. Xu, and J. Yu, “Chemical reaction
Algorithms in Combinatorial Optimization: A Survey and optimization for the optimal power flow problem”. The 2012 IEEE
Classification”. Artificial Intelligence and Knowledge Engineering Congress on Evolutionary Computation (CEC 2012), Brisbane,
Applications: A Bioinspired Approach (pp” 41-53). Berlin, Heidelberg: Australia, 10-15 June 2012. In IEEE CEC Proceedings, 2012, pp: 1-8,
Springer,2005, https://doi.org/10.1007/11499305_5. https://doi.org/10.1109/CEC.2012.6253003.
[26] V. Cung, S. Martins, C. Ribeiro, and C. Roucairol. “Strategies for the [41] Y. Khanafseh, M. Surakhi, A. Sharieh, and A. Sleit, “A Comparison
Parallel Implementation of Metaheuristics”. Essays and Surveys in between Chemical Reaction Optimization and Genetic Algorithms for
Metaheuristics, US: Springer,2002, (pp. 263-308). Max Flow Problem”, International Journal of Advanced Computer
https://doi.org/10.1007/978-1-4615-1507-4_13. Science and Applications (IJACSA), 2017, 8(8), pp: 8-15,
[27] H. Kim, H. Lam, and S. Kang, “Chemical Reaction Optimization for http://dx.doi.org/10.14569/IJACSA.2017.080802.
Task Scheduling in Grid Computing”. IEEE Transactions on Parallel [42] R. Barham, A. Sharieh, and A. Sliet, “Chemical Reaction Optimization
and Distributed,2011, 22(10), pp: 1624 – 1631, for Max Flow Problem”, (IJACSA) International Journal of Advanced
https://doi.org/10.1109/TPDS.2011.35. Computer Science and Applications, 2016, 7(8), pp: 189-196.
[28] E. Loiola, N. de Abreu, P. Boaventura-Netto, P. Hahn, and T. Querido, [43] K. Xu,“Vertex Cover Benchmark Instances (DIMACS & BHOSLIB)”.
“A survey for the quadratic assignment problem”. Eur J Oper Res, 2007, IJEA (international journal of Experimental algorithms),2012, 3(1), pp:
176(2), pp:657–690, https://doi.org/10.1016/j.ejor.2005.09.032. 1-18.
[29] J. Xu, A. Lam, and V. Li, “Parallel Chemical Reaction Optimization for [44] Penn State Harrisburg University. Vertex Cover Benchmark Instances,
the Quadratic Assignment Problem”. Proceedings of the 2010 2019. [online] Available at: https://turing.cs.hbg.psu.edu/benchmarks/
International Conference on Genetic and Evolutionary Methods, GEM vertex_cover.html [Accessed 27 March 2019].
2010, July 12-15, 2010, Las Vegas Nevada, USA. [45] DIMACS. the Center for Discrete Mathematics and Theoretical
[30] E. Demeulemeester and W. Herroelen,“Project scheduling: a research Computer Science, 2019. [online] Available at: http://dimacs.rutgers.edu
handbook”. Academic Publishers, Boston, MA, USA, 2002, [Accessed 8 March 2019].
https://doi.org/10.1007/b101924. [46] A. Al-Adwan, B. Mahafzah, and A. Sharieh, “Solving traveling
[31] A. Subramanian, H. Gupta, S. Das, and J. Cao, “Minimum interference salesman problem using parallel repetitive nearest neighbor algorithm on
channel assignment in multiradio wireless mesh networks”. IEEE Trans OTIS-Hypercube and OTIS-Mesh optoelectronic architectures”, Journal
Mobile Comput, 2008, 7(12), pp:1459–1473, of Supercomputing, 2018, 74(1), pp: 1-36,
https://doi.org/10.1109/TMC.2008.70. https://doi.org/10.1007/s11227-017-2102-y.
[32] A. Lam, J. Xu, and V. Li, “Chemical reaction optimization for [47] A. Al-Adwan, A. Sharieh, and B. Mahafzah, "Parallel heuristic local
population transition in peer-to-peer live streaming”. Proceedings of the search algorithm on OTIS hyper hexa-cell and OTIS mesh of trees
IEEE congress on evolutionary computation. Barcelona, Spain, 2010, optoelectronic architectures" Applied Intelligence, 2018, 49(10), pp: 1-
https://doi.org/10.1109/CEC.2010.5585933. 28, https://doi.org/10.1007/s10489-018-1283-2.
[33] B. Pan, A. Lam, and V. Li, “Network coding optimization based on [48] B. Mahafzah, M. Alshraideh, T. Abu-Kabeer, E. Ahmad, and N. Hamad,
chemical reaction optimization”. Proceedings of the IEEE global “The optical chained-cubic tree interconnection network: Topological
communications conference. Houston, TX, USA, 2011, structure and properties” Computers & Electrical Engineering,2012,
https://doi.org/10.1109/GLOCOM.2011.6133697. 38(2), pp: 330-345, https://doi.org/10.1016/j.compeleceng.2011.11.023.
91 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Internet of Things (IoT) has been industrially of cash in equipment or software. As a matter of fact, the
investigated as Platforms as a Services (PaaS). The naive design execution of the cloud introduces a few related ideas. These
of these types of services is to join the classic centralized Cloud ideas manage virtualization, resource allocation, computing
computing infrastructure with IoT services. This joining is also distribution, utilization of bandwidth, load balancing, fault
called CoT (Cloud of Things). In spite of the increasing resource tolerance, high availability and dynamic scalability for various
utilization of cloud computing, but it faces different challenges classifications of data and applications. The administration of
such as high latency, network failure, resource limitations, fault the operations identified with every one of these concepts is
tolerance and security etc. In order to address these challenges, performed by the cloud service provider.
fog computing is used. Fog computing is an extension of the cloud
system, which provides closer resources to IoT devices. It is The cloud providers allocate the resources to the end clients
worth mentioning that the scheduling mechanisms of IoT services as a service relying upon the uniqueness of the service models
work as a pivotal function in resource allocation for the cloud, or and furthermore dependent on the client needs. The service
fog computing. The scheduling methods guarantee the high models may incorporate Software as a service known as
availability and maximize utilization of the system resources. SAAS, Platform as a service known as PAAS, Infrastructure as
Most of the previous scheduling methods are based on a service known as IAAS. These services are inclined on one
centralized scheduling node, which represents a bottleneck for another and in a pool way.
the system. In this paper, we propose a new scheduling model for
manage real time and soft service requests in Fog systems, which By and large, the executions of the various procedures on
is called Decentralize Load-Balance Scheduling (DLBS). The the cloud present a few advantages to the end clients. At First,
proposed model provides decentralized load balancing control the data is shared more than one stage, so better services are
algorithm. This model distributes the load based on the type of conveyed to every user. Also, the end user can get the services
the service requests and the load status of each fog node. resources on-demand, flexible, reliable and portable way as
Moreover, this model spreads the load between system nodes like indicated by his need as it were.
wind flow, it migrates the tasks from the high load node to the
closest low load node. Hence the load is expanded overall the In spite of these advantages that can be offered by cloud
system dynamically. Finally, The DLBS is simulated and computing to enormous applications, it faces a lot of
evaluated on truthful fog environment. challenges [3]. The first challenge happens when the number of
the clients is increased. For this situation, the requests are
Keywords—Cloud computing; fog computing; mist computing; broadened to increase the number of services than the cloud
IoT; load balancing; reliability capacities. As client requests is increased, as the responses time
is increased unless the available resources and the available
I. INTRODUCTION
bandwidth are upraised to acquire all the extra requests. The
Cloud computing is presented as an ongoing innovation, second challenge happens when the created data by the cloud
which is totally dependent on the web. The engineering of the services is migrated through a long distance from the cloud to
Cloud computing depends on a focal server that keep up a the clients. The far distance creates additional challenge about
tremendous measure of sharing database, various assets and an the data security. Moreover, an unpredictable abundance in the
enormous number of business applications. Then again, a workload may cause the need to create a novel load balancing
colossal number of remote customers that has a place with strategy. The load balancing is the reasonable assignment of
various associations can profit by the various administrations the task among the parallel resources such as networking, hard
given by the focal server. Every remote client has its own, drives and computers [4]. In this way, it will be required to
working framework and internet browser that work achieve the improvement in the distribution of the computation
autonomously on the substance of the cloud server [1, 2]. The resources and storage devices. So as to beat these challenges,
association of the client to the web is the main prerequisite another innovation of profoundly virtualized processing model
from the client to use the cloud server capacities. Along these has been displayed known as Fog computing. The model [5] is
lines, the IT business and any little association can get these proposed by CISCO to be held as cloud edge of an enterprise
services from the cloud without spending tremendous measure
92 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
93 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In this area, the first load balancing technique is introduced with the minimum turnaround time and allocate the task to the
in [15]. This method is intended to achieve good services by processor core that can minimize its turnaround time, repeating
increasing the resource utilization based on two parameters, the process many times until all tasks are allocated. The same
which are the task priority and its length. The choice of the as the MCT algorithm, the MIN-MIN algorithm is also based
tasks for the scheduling might be gotten from both of the first on the minimum completion time. The MIN-MIN algorithm
and last indexed queue to accomplish an all the more relentless proposes all tasks that are not scheduled, but the MCT
framework. algorithm considers unique task at a time. The MAX-MIN
algorithm is similar to the MIN-MIN algorithm, which also
The tasks are scheduled dependent on the total credit computes minimum completion time without scheduled tasks
system sponsored from grouping of credit length computed firstly and then selects the task with the largest minimum
from task length and credit priority computed from the task completion time and assigns the task to the processor core with
priority. Finally, the priority of processing is given to the high the minimum completion time.
credit task. However, this algorithm suffers from certain
shortcomings when the absolute credits of several tasks Mondala et at. use an optimized approach algorithm to have
became indistinguishable. For this situation, the FCFS has to load balancing scheduling system [20]. This model is based on
be added without guarantee of tasks to be completed earlier or a centralized load balancing algorithm. In another words, the
to its deadline. system is based on a central node that distributes the workload
tasks. Hence, the main drawback is of this model is that if the
Another algorithm depends on comparable to conduct of central node fails, the whole working of the system will fail.
honey bee model (HBB-LB) is proposed by Dhinesh babu L.D This means that the central node is represent the system
et al. [16]. In this algorithm, the priority is taken as a bottleneck. So here, using decentralized load balancing strategy
fundamental QoS factor to Bar any procedure from hanging solves this bottleneck. Resource utilization can be done
tight for quite a while in the line to diminish the execution time effectively to enhance the throughput, accordingly decreasing
and augment the throughput. Similarly, the tasks can be acted the cost of an application running in a SAAS environment
as the Honey bees and the Virtual Machines can be acted as
without break service level agreements [21].
sustenance sources. Moreover, The VMs are classified
according to three circumstances, balanced overload, high Actually, the different scheduling algorithms based on QoS
overload and low overload. When the VMs are overloaded, the parameters have been introduced for different environments in
tasks are evacuated and act as a honey bee. So, these tasks are [22]. The scheduling is performed to achieve the huge service
migrated to the low load VMs. These duties are depending on requests and to enhance the efficiency of the workload.
how many high priority tasks are executed on those VMs. It Subsequently, there are numerous modules that are
should be noticed that the VM is chosen based on the low implemented in each kind of the scheduling algorithms, for
overload and the least number of the executed priority tasks. example, Min-Min, FCFS, Max-Min, Round-Robin algorithm.
After proper tasks on VM, data is refreshed with the goal that
the rest of the assignments can acquire their needs under load Nevertheless, the one of the efficient methods among them
VM. This algorithm has presented certain advantages is the heuristic method. Its allocating the tasks includes three
represented in the proper resource utilization; maximizing the stages in a cloud computing. At first, the VMs are located.
throughput while keeping different QOS parameters which are Hence, the best target VM is chosen. At last, the task is
built on the task priority. On the other hand, the disadvantages assigned to the target VM. Lately, the Real Efficient Time
are introduced for the low need priority tasks which suffer Scheduling (RETS) is investigated in [23]. The main goal of
from idle state or long time waiting in the queue. These tasks RETS is to process the real-time tasks without delay.
may be dismissed causing the unbalancing of the workload Therefore, it keeps one tenth of the available resources for the
balancing. real-time tasks. Although, this ratio can be insufficient if the
real-time tasks exceed this ratio. On the other hand, one tenth
For an enormous scale condition, e.g., cloud computing of the available resources will be idle if there are no real-time
framework, there had been also various scheduling approaches tasks.
proposed with the objective of accomplishing the better task
execution time for cloud resources [17]. Independent task Moreover, Anju et al. introduces multilevel of priority-
scheduling algorithms mainly include MCT algorithm [18], based task scheduling algorithm (PBATS)[24, 25]. This
MET algorithm [15], MIN-MIN algorithm [15], MAX-MIN algorithm has three levels of priorities, which prioritizes the
algorithm [19], PMM algorithm, and genetic algorithm. The tasks based on the length of the instructions. Also, to enhance
MCT (Minimum Completion Time) algorithm assigns each performance of PBATS, it migrates the tasks under the
task in any order to the processor core that causes the task to be minimum migration time policy. This policy can cause
completed at the earliest time. It prohibits some tasks to be overload of node, which has low network overhead. Also, this
allocated to the fastest processor core. The MET (Minimum policy doesn't distinguish between the real and soft tasks.
Execution Time) algorithm allocates each task to a processor Also, Wang et al. proposed a task scheduling algorithm in
core in any order that minify the task execution time. As the fog computing, which is called “hybrid heuristic (HH)”
opposed to the MCT algorithm, the MET algorithm does not algorithm [26]. HH algorithm is mainly focus in solving high
consider the processor core’s ready time, which may prompt energy consumption in case of using limited computing
genuine burden unevenness crosswise over processor cores. resources. Unfortunately, HH method isn’t distinguish between
The MIN-MIN algorithms calculates the minimum completion the mist and middle fog nods. Hence, this algorithm is not
time of all unscheduled tasks firstly, and then chooses the task efficient method for real-time services.
94 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
III. PROPOSED MODEL Mist node gives the real time task queue in resources
In a Fog computing environment, the load balancing is a allocation. The tasks in the real time queue will be allocated
pivot point for effective and efficient resource utilization, into one of idle local VMs in the node. If there is no idle, LBA
bandwidth and to achieves desired quality of service (QoS). preempt one of soft task VMs. In the worst case scenario, if
Fog Computing system is divided virtually in two type of there are no idle or soft VMs, Fog explorer module suggest the
nodes, namely; mist and middle edge node. Actually, both resources in the closest mist/middle edge node. Fog explorer
types of fog nodes can have the same structure and resources. detects the status of the other fog node by getting the status
Nevertheless, the most closed node to IoT is called mist. Each flags. The status flags are set by LBA module and broadcasted
Mist computing server is centered in the specific location by the fog explorer. Each Mist node has four types of status
mainly to receive the clients or/and IoT requests in a specific flags, which determine the status of the node, namely, load
region. The fog colony is connected to a cloud system in the lock, real task lock, receive status, and send status. Load lock
case of fog resources shortage to overcome the fulfillment of flag, which is soft task waiting, is set by zero if the expected
task requests. waiting time will not exceed QoS threshold (λ). In another
word, λ grantees that the service of the soft tasks will be
In this paper, the new scheduling model (DLBS) is provided in a reasonable delay. If load lock flag is set by one,
proposed in the cloud-fog-mist environment. The structure of this fog node can't receive a soft task from other fog and its soft
this model is shown in Fig. 2. First of all, the Service Listener tasks will migrate outside the node. Also, real-time task lock is
(SL) receives the user/IoT service request. Hence, SL creates a set by one if all VMs are allocated by real-time tasks. For any
task for the service request and sends it to Load Balancing fog node if one of VMs is processing a soft task, the real-time
Allocator (LBA) module with required software from service task lock is set by zero. Finally, according to task migration the
container. Also, SL send task-metadata like, task type (real fog node blocks the receiving tasks from other nodes if its
time or soft), expected execution time, etc. So, each Mist server receive status or send status has value one. Obviously, the
is supplied by its own Load Balancing Allocator module status flags are used to maintain the system balanced and
(LBA). LBA is responsible for allocating the clients and/or IoT available.
service requests into the fog resources. There two types of
user/IoT request; real time and soft-tasks. The proposed model Example in Fig. 3 shows the closer fog region for Mist Y
is designed to handle both types of tasks. by dotted line, and the closer region for middle edge node C by
the dashed line. In this example, Mist Y receives two service
Mist node gives the real time task queue in resources requests from IoT devices. The first request is real- time
allocation. The tasks in the real time queue will be allocated request, which come from Pacemaker device. This type of
into one of idle local VMs in the node. If there is no idle, LBA request is classified by the Fog Explorer as real-time request.
preempt one of soft task VMs. In the worst case scenario, if Hence, this request must be handled in the local fog (nod Y).
there are no idle or soft VMs, Fog explorer module suggest the On the contrary, Mist Y is forwarding the soft request to
resources in the closest mist/middle edge node. Fog explorer middle edge server C. Also, for the node C the load is
detects the status of the other fog node by getting the status migrating to D. This strategy makes the load spread over all
flags. The status flags are set by LBA module and broadcasted system nodes.
by the fog explorer. Each Mist node has four types of status
flags, which determine the status of the node, namely, load
lock, real task lock, receive status, and send status. Load lock
flag, which is soft task waiting, is set by zero if the expected
waiting time will not exceed QoS threshold ( λ). In another
word, λ grantees that the service of the soft tasks will be
provided in a reasonable delay. If load lock flag is set by one,
this fog node can't receive a soft task from other fog and its soft
tasks will migrate outside the node. Also, real-time task lock is
set by one if all VMs are allocated by real-time tasks. For any
fog node if one of VMs is processing a soft task, the real-time
task lock is set by zero. Finally, according to task migration the
fog node blocks the receiving tasks from other nodes if its
receive status or send status has value one. Obviously, the
status flags are used to maintain the system balanced and
available.
Fig. 2. DLBS Model.
95 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
96 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
34.End if namely; turnaround time, the average waiting time and the
throughput. Finally, the third diminution measures the
B. Fog Explorer, Service Container and Flags suitability of the model for the real-time services by evaluating
Finally, fog explorer module is responsible for determine the number of failed tasks in the compared algorithms.
the closer fog region for each node. This region is defined as
set of nodes which has minimum communication overhead. If This section is organized into three subsections. Each
any of the status flags is change in each node, the node subsection is concerned to evaluate a performance dimension.
broadcast this information to its closed region. Also, Fog Hence, the following subsection evaluates the performance of
explorer is responsible for broadcasting a copy of the Service the system using all types of tasks. Moreover, the second
Container to all fog and mist computing nodes. Moreover, it subsection evaluates the effect of the system on the real time
should send up-to-date a copy of additional changes in Service tasks only. Finally, the failure in the real-service requests is
Container. measured in third Subsection.
A. System Performance using Real-Time and Soft Service
IV. SIMULATION SETUP
Requests
As a mist landscape, we propose a fog-mist colony of 100 In this section three tests are done. The first test measures
nodes. Half of the colony nodes are mist nodes, which receive the response time of variant number of tasks. The second test
the user requests. The fog-mist colony is connected to a cloud evaluate the waiting time of the system. Finally, the last test in
system in circumstance of shortage in the fog-mist sources to this section measures the throughput.
the fulfillment of soft task requests. Of these 100 fog-mist
colony, 10 are concurrently issuing 1,000 task requests to the 1) Turnaround time performance test: The first experiment
mist colony. Furthermore, IoT applications are characterized measure the system performance based on the Turnaround time
by two types (real and soft). parameter. DLBS is compared with previous mentioned four
The proposed DLBS algorithm, have been implemented on algorithms. The experiments are done using different number
simulator CloudSim [27, 28] 3.0.2 to execute tasks along with of workloads from 1000 to 10,000 tasks. The real time tasks
Window 7 OS, core i5 2.3 GHz processor and NetBeans IDE will represent 20% from all of the inserted workload in each
7.2.1. CloudSim computes the execution time of a service experiment. Obviously, we can notice that the FCFS curve is
request to fulfill a task requirement, hence computes the rapidly increased by increasing the number of service requests.
waiting time for soft task by aggregating the number of The bad performance of FCFS is due to the non-preemptive
instructions necessary to execute the waiting soft tasks. In this property. Also, Max-Min curve is closed to the FCFS curve.
experiment, the soft-task request and real-task requests
Since, the Max-Min is allocating the longest tasks to VMs
required 0.05, and 0.03 million of instructions per second
(mips) respectively. Both task types have 300 MB of incoming which has lest remaining execution time. In another word, in
and 300 MB of outgoing data. Fog/Mist nodes able to able to Max-Min scheduling algorithms the short tasks will wait a long
handle 250 MIPS. Each fog node can create 10 VM’s have the time to get the resources, which increase the average of waiting
processing power 500 MIPS. The bandwidth between fog time. In addition, the PBATS curve is keep a less in the
nodes is set to 100 Mbit/s, and between the cloud and fog average turnaround time results when compared to the FCFS
nodes to 10 Mbit/s. All experiments are repeated for 10 times and Max-Min. Indeed, the tasks in the PBATS algorithm are
and the mean values are taken. classified into three levels of priorities and underestimate the
DLBS model is compared with four models. The first quality of services. Furthermore, the curve of the RETS refer to
model is FCFS, which serve the tasks based the arrival time. acceptable results with a light load up to 1,500 tasks, as shown
Moreover, the others compared models was created for the in Fig. 4(A). Also, RETS gives an inefficient performance if
cloud computing system, namely the Max-Min, the PBATS compared by the proposed algorithm (DLBS). The
and the RETS. The Max-Min maintains a task status table to performance of RETS is decreased as increasing the work load.
envision the real loads of the VMs and the evaluated finishing The performance deterioration of RETS algorithm is due to
time of tasks, which can distribute the workload among nodes static reservation for the real tasks. It assign one tenth of the
[29]. The Priority Based Autonomic Task Scheduling (PBATS) resources for the real requests. Reserving a static ratio of the
that schedule its tasks according to three different priorities
resources can cause problem if there are no proper real tasks.
levels [25, 30]. Furthermore, the Real Time Efficient
Scheduling (RETS) depends on reserving a one tenth of the Actually, it is a dilemma if the real tasks exceed the reserved
resources for the real-tasks [23]. All these scheduling resources. Actually, the DLBS overcome these problems. It
techniques are matched by the proposed techniques to evaluate gives high priority to the real tasks for satisfy its deadline.
the load balancing in the proposed model. Also, it maintains a specific response time for the soft tasks.
Subsequently, the DLBS is the most efficient algorithm among
V. RESULTS AND DISCUSION all of the compared algorithms in the Mist-fog environment.
The performance evaluations have been performed in three 2) The waiting time performance test: This experiment
dimensions. The first dimension evaluates the performance of measures the waiting time for the service request tasks. As
the system on the soft-tasks load. On another hand, the second shown in Fig. 4(B), the waiting time of the DLBS curve has the
dimension measures the system reliability for the real-time best performance. Moreover, for having a certain QoS the
tasks. The performance evaluation based on three parameters, expected waiting time for the soft tasks parameter λ is set by 10
97 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
second. Hence, the DLBS curve values are very close to ten time service according to their deadline. Actually, these
second after 5,000 tasks. It is worth noting that the FCFS curve algorithms are not indeed to handle the real-time tasks. Hence,
has the worst performance. This bad performance is caused by the real times tasks are treated as the soft tasks. On other hand,
the same reasons that increase the average turnaround times. the RETS gives an acceptable performance when the number
Also, the Max-Min curve is the closest one to the FCFS curve. of the Real-Tasks are not exceed one tenth of the system
In the PBATS curve the tasks allocation is depending on three resources. As obtained the figure, the performance of RETS
levels of priorities, which increase the waiting time for tasks result is acceptable until 1,000 real-tasks and is decay after this
according to their levels. Furthermore, the RETS algorithm has point. Furthermore, the RETS algorithm preserve the response
an acceptable performance until the workload less than or time of the real time tasks to be less than their deadline times.
equal to 3,000 tasks. Unfortunately, as increasing the services In other words, the real-time tasks are not presented to any
requests, as the average of waiting time is rapidly increased for postpone, which limits the turnaround time. Moreover, the real-
the RETS. All of these problems have been solved by DLBS time tasks are migrated from fog node to another one to avoid
algorithm as shown by the performance curve. DLBS waiting time.
maintains an upper bound of the waiting time for each soft task 2) The waiting time performance test: The averages of
in mist node and send the exceeding load to the closest low waiting time curves that expose the impact of the DLBS model
load node or to the middle edge node. on the duration time of the real tasks are shown in Fig. 5(B). In
3) The throughput performance test: This test measure the this figure, the lower mean waiting time is implied for DLBS.
performance based on the average of system throughput. The As mentioned before, the DLBS model is designed to give the
throughput is defined as the total number of finished tasks per first priority for the Real-Time tasks. Hence, the reserved
time. Additionally, the experiment is done based on the same resources for the soft tasks are released to allocate the real
workload of the past examination. The performance of the tasks. However, the RETS is the closest curve among all the
compared algorithms is shown in Fig. 4(C). We can notice that, compared algorithms to the DLBS. Unfortunately, as the real
the throughput of DLBS has the best throughputs enhancement requests load in RETS is increased, as the average waiting time
compared by the other algorithms. The performance is increased. Hence, the deadline times of the real tasks will be
enhancement of DLBS is caused by the balanced distribution exceeded in RETS model.
of the tasks that satisfy QoS. Also, the worst performance 3) The throughput performance test: The throughputs
curve is the FCFS. Moreover, the RETS throughput curve is curves, in Fig. 5(C), show the performance comparison
successor to DLBS curve. Since RETS gives the highest between the competitive algorithms. Unmistakably, the highest
priority to the real tasks, which is the lightest processing tasks, throughput is accomplished by DLBS. The RETS throughput
then it increases the number of the finished tasks. becomes consistent after satisfying the reserved ratio of the real
tasks. Also, the DLBS throughput is increased as increasing the
B. System Performance using Real-Time Service Requests
real time tasks. Since DLBS algorithm can assign the whole
This experiment evaluates the effect of the proposed system mist node resources and borrows additional resources to satisfy
on the real time tasks compared with other algorithms. Each
the real-time service requests. Moreover, FCFS has the worst
experiment is completed on the real task ratio 25% of
workload. Through the experiments, the workloads for all the performance because it isn't careless about the deadline of the
tasks types are changed from 1,000 to 10,000 tasks. Hence, the real-time task.
real time tasks are changed from 250 to 2,500 tasks. However, 4) Real-time task failure test: To judge about the
all the experiments of the Real-Time tasks are performed in the suitability of the algorithm for real time services, the task
existence of the soft tasks load. This section is organized as failure should be concerned. To judge about the suitability of
follows. The following subsection measures the turnaround the algorithm for real time services, the task failure should be
time. Subsection (2) measures the waiting time and Subsection considered. Fig. 5(D) measures real task failure for the
(3) measures the throughput. Finally, the Subsection (4) proposed and the compared algorithms. The number of task
measures the suitability of the system for real service. failure for the DLBS model is trivial if compared with the other
1) Turnaround Time performance Test: The turnaround models. RETS model gives a good performance in low load of
time performance comparison of the compared algorithms is the real time tasks. Unfortunately, RETS model doesn't
shown in Fig. 5(A). The worst performance is obtained by the supports flexibility in the reserved resources for the real time
curves that represent the FCFS, PBATS and the Max-Min tasks. Also, it doesn't support task migration to provide the
algorithms respectively. The essential shortage of these desired resources. The other algorithms failure values indicate
algorithms is the disability to handle the requests of the real inaptitude for real time services.
98 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
99 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
VI. CONCLUSION AND FUTURE WORK environment”, International Journal of Computer Science and
Communication, Vol. 8 • Number 2 Jan - Jun 2015 pp. 43-49.
In this paper, DLBS model is designed for managing soft [14] S. F. El-Zoghdy and S. Ghoniemy, “A Survey of Load Balancing In
and real time services in fog computing environments. The High-Performance Distributed Computing Systems”, International
DLBS model introduces decentralize scheduling algorithm. Journal of Advanced Computing Research, Volume 1, 2014.
Fog computing consists of two type of nodes, namely; mist and [15] Mohsen and Hossein Delda, “Balancing Load in a Computational Grid
middle edge nodes. Mist nodes are the closer nodes to IoT Applying Adaptive, Intelligent Colonies of Ants”, Informatica 32 (2008)
devise, which receive its services requests. The DLBS model 327–335.
provides an efficient solution for having IoT service response [16] Brototi Mondala, Kousik Dasguptaa, Paramartha Duttab”Load
Balancing in Cloud Computing using Stochastic Hill Climbing-A Soft
time. This model is providing an efficient load balancing Computing Approach”, Elsevier, Procedia Technology 4(2012) pp. 783
strategy for IoT service requests. Also, this model manages the –789.
IoT services requests load for each fog node in decentralize [17] W. Lin, C. Zhu, J. Li, B. Liu, and H. Lian, “Novel algorithms and
manner. The decentralize load management avoids the equivalence optimisation for resource allocation in cloud
bottleneck problem, which exists in the majority of the other computing,”International Journal of Web and Grid Services,vol. 11, no.
solution. Moreover, this model is designed to fit the real-time 2, pp. 69–78, 2015.
serves requests. The experiments show that our methods [18] M.Maheswaran,S.Ali,H.J.Siegel,D.Hensgen,andR.F. Freund, “Dynamic
mapping of a class of independent tasks onto heterogeneous computing
outperform the compared methods. In future work, this model systems,”Journal of Parallel and Distributed Computing, vol. 59, no. 2,
will be developed to manage the heterogeneous Mist nodes. pp. 107–131, 1999.
REFERENCES [19] T.D.Brauny,H.Siegely,N.Beckyetal.,“AComparison Study of Static
[1] Chandrasekhar S. Pawar, Rajnikant B. Wagh, Priority Based Dynamic Mapping Heuristics for a Class of Meta-tasks on Heterogeneous
resource allocation in Cloud Computing with modified Waiting Queue, Computing Systems,”parallel & distributed computing, vol.61, no.6,
Proceeding of the IEEE 2013 International Conference on Intelligent pp.810–837,2001.
System and Signal Processing(ISSP) Pages 311-316. [20] Brototi Mondala, Kousik Dasguptaa, Paramartha Duttab”Load
[2] Yusen Li, Xueyan Tang, Wentong Cai, Dynamic Bin packing for on Balancing in Cloud Computing using Stochastic Hill Climbing-A Soft
demand cloud resource allocation, Proceedings of the IEEE Transactions Computing Approach”, Elsevier, Procedia Technology 4(2012) pp. 783
on Parallel and Distributed Systems ,2015,Paged 1-14. –789.
[3] Savani Nirav M, Prof. Amar Buchade,―Priority Based Allocation in [21] Atul Vikas Luthra and Dharmendra Kumar Yadav,”MultiObjective
Cloud Computing, International Journal of Engineering Research & Tasks Scheduling Algorithm for Cloud Computing Throughput
Technology (IJERT) ISSN: 2278-0181 IJERTV3IS051140 Vol. 3 Issue Optimization”, International Conference on Intelligent, Communication
5, May – 2014. & Convergence, Procedia Computer Science 48(2015) 107- 113.
[4] Brototi Mondala, Kousik Dasguptaa, Paramartha Duttab”Load [22] Mohamed A. Elsharkawey, Hosam E. Refaat,"CVSHR: Enchantment
Balancing in Cloud Computing using Stochastic Hill Climbing-A Soft Cloud-based Video Streaming using the Heterogeneous Resource
Computing Approach”, Elsevier, Procedia Technology 4(2012) pp. 783 Allocation", International Journal of Computer Network and
–789. Information Security (IJCNIS), Vol.9, No.9, pp.1-11, 2017.DOI:
10.5815/ijcnis.2017.09.01.
[5] Ivan Stojmenovic, sheng Wen, “The Fog Computing Paradigm:
Scenarios and security issues” Proceedings of the IEEE International [23] M.Verma, N. Bhardwaj and A. Kumar, "Real Time Efficient Scheduling
Fedrerated Conference on Computer Science and Information Systems, Algorithm for Load Balancing in Fog Computing Environment",I.J.
2014, pp.1-8. Information Technology and Computer Science, April, 2016, 4, 1-10.
[6] Mahmood A., Zen H. (2018) Toward Edge-based Caching in Software- [24] B.Anju and C.Inderveer (2016), "Multilevel Priority-Based Task
defined Heterogeneous Vehicular Networks. In: Mahmood Z. (eds) Fog Scheduling Algorithm for Workflows in Cloud Computing
Computing. Springer, Cham. https://doi.org/10.1007/978-3-319-94890- Environment". In Proceedings of International Conference on ICT for
4_13. Sustainable Development: Volume.
[7] Sari A. (2018). Context-Aware Intelligent Systems for Fog Computing [25] Swati Agarwal, Shashank Yadav, Arun Kumar Yadav,"An Efficient
Environments for Cyber-Threat Intelligence. In Fog Computing (pp. Architecture and Algorithm for Resource Provisioning in Fog
205–225). Cham: Springer. 10.1007/978-3-319-94890-4_10. Computing", International Journal of Information Engineering and
Electronic Business(IJIEEB), Vol.8, No.1, pp.48-61, 2016. DOI:
[8] Nanxi Chen, Yang Yang, Tao Zhang, Ming-Tuo Zhou, Xiliang Luo, 10.5815/ijieeb.2016.01.06.
John K. Zao, "Fog as a Service Technology", Communications
Magazine IEEE, vol. 56, no. 11, pp. 95-101, 2018. [26] Wang J, Li D. Task Scheduling Based on a Hybrid Heuristic Algorithm
for Smart Production Line with Fog Computing. Sensors (Basel).
[9] Luthra, M., Koldehofe, B. & Steinmetz, R. "Transitions for Increased 2019;19(5):1023. Published 2019 Feb 28. doi:10.3390/s19051023.
Flexibility in Fog Computing: A Case Study on Complex Event
Processing" Informatik Spektrum (2019). [27] W. Chen and E. Deelman,―Workflowsim: A toolkit for simulating
https://doi.org/10.1007/s00287-019-01191-0. scientific workflows in distributed environments, in 2012 IEEE 8th
International Conference on E-Science, ser. eScience, 2012, pp. 1–8.
[10] [7] A. Davies, Cisco pushes IoT analytics to the extreme edge with mist [Online]. Available:https://github.com/WorkflowSim.
computing. [Online]. Available: http://rethinkresearch.biz/articles/cisco-
pushes-iotanalytics-extreme-edge-mist-computing-2, Blog, Rethink [28] R. N. Calheiros, R. Ranjan, A. Beloglazov, C. A. F. De Rose, and R.
Research. Buyya,―CloudSim: a toolkit for modeling and simulation of cloud
computing environments and evaluation of resource provisioning
[11] J.S.Preden, K.Tammemäe, A.Jantsch, M.Leier, A.Riid, E.Calis, The algorithms, Software: Practice and Experience, vol. 41, no. 1, 2011.
benefits of self-awareness and attention in fog and mist computing,
Comput. (Long Beach Calif) 48(7)(2015)37–45. [29] Xiaofang Li, Yingchi Mao, Xianjian Xiao, "An improved Max-Min
task-scheduling algorithm for elastic cloud", Computer, Consumer and
[12] Flavio Bonomi, Rodolfo Milito, Jiang Zhu, Sateesh Addepalli “Fog Control (IS3C), 2014 International Symposium on.
Computing and its Role in the internet of things”,
http://conferences.sigcomm.org/sigcomm/2012/pa per/mcc/p13.pdf. [30] B.Anju and C.Inderveer (2016), "Multilevel Priority-Based Task
Scheduling Algorithm for Workflows in Cloud Computing
[13] Manisha Verma, Neelam Bhardwaj Arun Kumar Yadav,” An Environment". In Proceedings of International Conference on ICT for
architecture for load balancing techniques for Fog computing Sustainable Development: Volume.
100 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In the current economic realities, the forms of The survival of industrial enterprises in conditions of
integration business entities through the creation of enterprise economic instability often directly depends on a successful
groups (EGs), reorganized from industry structures or created a management of business processes. At the same time,
new by acquiring existing companies, are becoming increasingly considerable difficulties appear in the EGs already at the
relevant. The economic activity of the enterprise is carried out in initial stage of modernization of management [6]. This is due
the conditions of economic instability and improvement of the to the fact that most of the existing traditional software for
system economic relations, which imposes fundamentally new managing the development of enterprises are built on the
requirements in the sphere of managing the interaction of classical principles of budgeting - control and are not
enterprises. Under these conditions, the successful development
sufficiently effective at present to manage the business
of the enterprises and often their very existence depend both on
processes of EGs. In addition, the transition from the
the effective use of the management systems themselves and on
the competence of the management decisions made. designated strategy for the development of EGs to specific
Consequently, for decision makers and managers of Group actions of the performers is sometimes difficult to implement
Policy (GP), the problem of evaluating the development of GP due to the lack of a regular unified mechanism that would set
and promptly making sound management decisions in an the necessary priorities, allow preparing and evaluating
unstable and rapidly changing economic environment is solutions, analyzing the distribution of IP across EGs [7], and
considered a particular relevance. One of the promising ways to controlling the consistency and efficiency of execution in IP
solve this problem is the development of decision support systems and also supported the possibility of joint decision-making on
(DSS), using scientifically based decision-making methods based a number of current issues [8].
on modern mathematical apparatus and computer equipment. At
present, the approach to managing the development of the EGs is However, it should be noted that the state of affairs in the
associated with the representation of the latter as a multi-agent field of DSS applications to ensure the effectiveness of
system (MAS). The DSS does not replace, but complements the business process management does not sufficiently meet the
existing management systems in the EGs, interacting with them, needs of the enterprise in modern conditions, and there are a
and uses in its work information about the functioning of EGs number of scientific problems that require systemic solutions
units. [9]. Among them, it can be noted: the need to develop a
decision-making methodology in determining the optimal
Keywords—Management systems; decision support systems; control regimes for the interaction of industrial enterprises in
multi-agent systems; group policy; enterprise groups the EGs; the need to develop the principles of information and
analytical support for DSS when optimizing the management
I. INTRODUCTION systems of a EGs [10], the lack of common models for
At all stages of the economic development of EGs, the selecting and implementing IPs in several EGs under the
most important problem in the activities of enterprises is the direction of the managing company (MC). Therefore, the
problem of increasing the competitiveness of their products, development and applications of DSS in the management of
which can be achieved both by modernizing production and business processes and EGs is currently a pressing and timely
by optimizing the interaction of enterprises in EGs [1],[2]. task.
One of the priorities is the development and use of DSS in the
The aim of the work is a systematic analysis of
management of business processes EGs.
information processes for managing the activities of EGs, the
Under the management of business processes we can development of DSS to increase the efficiency of the
smoothly understand the system of targeted impacts, in which, implementation in IP, as well as the optimization of teamwork
by means of decision-makers, decision makers implement (interaction) of enterprises belonging to the EGs. To achieve
measures to improve the efficiency of the EGs. Several types these goals, the main research methods used were: system
of impacts in this research are considered: the selection and analysis, automatic control theory, decision theory, structural
implementation of investment projects (IP) [3], ensuring the and dynamic analysis, system modeling, numerical methods,
production of competitive products; optimizations of the nonlinear programming [9].
system parameters interaction with agents in MAS for the
Practical values of this research are considered the
manufacture and sale of products; restructuring of EGs units
applications of the developed models in the management
[4],[5].
system of joint business processes in EGs allows taking into
account the change in the performance indicators of each
101 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
industrial enterprises over time of their joint work [7],[11]. popular information processing systems, their functionality,
This makes it possible to conduct an analysis of the stability in architecture, software platforms, design principles, etc. is
the functioning of EGs and to take control decisions at the considered. The analysis of preliminary data preparation
optimal points in time to obtain maximum values of the systems for making management decisions based on ERP,
functional quality of each enterprise [12]. IDEF, MRP, ERPII standards was carried out. The advantages
and disadvantages of replicable systems of this class are noted,
II. RESEARCH FRAMEWORK including the complexity, duration and cost of
Modern concepts of managing complex distributed implementation, and examples of implemented DSS for
systems in various industries are based on the man-machine various fields of activity are given. It was concluded that there
organization of management processes, in which the role of are no systems applicable for the purposes of managing the
decision-making is assigned to a person, and the machine business processes of the EGs, taking into account the
provides information support for the stages of generating and optimization of the modes of cooperation between the MAS
generating alternative solutions [13]. It is noted that the use of agents [19],[20].
this approach is a necessary measure to overcome a high level In the next stage of this research, business process
of uncertainty in the formulation and conditions for solving management models are developed based on the selection and
problems of managing complex, nonlinear and dynamic implementation of IP in DSS. The proposed management
objects, such as EGs. At the same time, the effectiveness of models are conceptual in nature and reflect the most
the control systems of such EGs is largely determined by the significant aspects of the joint work of the EGs [6],[14]. In
subjective properties of the decision maker (DM) operating in this case, the main focus is on determining the basic laws of
the control loop, which in turn requires it to have a high level management processes, as well as the trends and strategies for
of competence both in managing business processes of the their development, depending on the parameters of the work
EGs and in the resulting problem situations [14],[15]. of the EGs between themselves and the external environment.
Currently known and widely used method of overcoming The analysis of the MAS works, consisting of industrial
subjectivity and increasing the level of competence of decision enterprises is under the control of the Criminal Code. It is
makers is the use of DSS. DSS provides the correctness of shown that each enterprise that is included in the EGs can be
solving problems by choosing rational options for managing represented as an independent agent acting within the limits of
business processes through the integrated use of a set of the powers assigned to it. From the position of system
technologically interconnected services that implement analysis, such MAS can be represented by several classes: the
traditional and advanced information technologies. central management company, territorial management
DSS can be represented as a set of management companies, industrial enterprises (agents A, B, C) (Fig. 1).
subsystems operating in the information environment in
accordance with established information flow procedures,
which determine the sequence of business process
implementation steps, as well as methods of transmitting,
storing and processing information in the management process
[16],[17]. Consequently, one of the main conditions for the
effective functioning of the DSS is a constant analysis and
processing, establishment of links and ways of circulating
information flows between information sources and receivers,
which provides an integrated approach to the development and
implementation of management decision options for decision
makers.
For management of enterprises known and well-
established in practice is an approach based on the use of
multi-agent DSS, which combines various methods of
analyzing situations and forming management decisions
implemented by agents of MAS [9]. It is shown that multi- Fig. 1. Scheme of Industrial Enterprises Group.
agent technologies can be considered as the basis for creating
effective methods and tools for developing business process III. PROPOSED METHODOLOGY AND EXPERIMENTAL
management systems. The issues of creation and application RESULTS
of DSS in situations related to the optimization of the In solving the problems of managing business processes of
interaction between agents in the network structures of the the implementation of IP, the Solow single-product model is
MAS and hierarchical structures are considered. The most used, which can be applied both to a separate industrial
important components of such systems, which include enterprise and to the entire EGs. In addition, it is assumed that
subsystems for modeling business processes, are highlighted the MAS operates in the established external environment,
[18]. whereas, there is no time lag between investments and their
An analytical review of the DSS structures and development, also the pace of enterprise development is
information processing methods in them is given. The most characterized by the dynamics of production assets, which in
102 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
turn is determined by the amount of investment resources while achieving the same values of Хт. Table I shows the
(deductions from profits or external investments). decrease in total investor costs (as a percentage) when
financing for D*(t) as compared to the rate of financing for
When optimizing the management of IP implementation at D0=const for a number of values of μ and λ for XT / X0=2 and
an industrial enterprise and the expense of external sources, it T=2.
is assumed that at the initial moment of time the level of the
fixed production assets of the enterprise in monetary terms is Thus, the management of the investment process according
X0, and in order to further develop the enterprise, it is to the obtained expression D*(t) compared to D0 = const can
required to increase it to the required value of XT due to reduce the total investor costs when the same values of XT are
foreign investment of the agent investor over time [0, T]. The reached.
dynamics of changes in the current level of cost for production
assets of the enterprise in this case can be written in the Also in this research discusses such IPs, when an industrial
following form: enterprise additionally uses its profits with selling an IP,
which it invests in investments or in its own savings. The
X ' (t ) X (t ) D(t ). (1)
choice of one of these options is made by the company from
the conditions for obtaining maximum profit. The investor
Here μ is the equipment depreciation factor, and D(t) is the participates in joint work by acquiring shares on which he
investment rate (the investor’s control parameter). An receives dividends. The equation of agents working together
investor, when participating in an IP, tries to control the in such MAS, by analogy with (1), will be:
parameter D(t) in such a way as to obtain the minimum values
of its own total investment costs G(T) (functional quality)
X ' (t ) X (t ) D(t ). (3)
T
G (T ) D(t )e t dt
min, Here π is the profit, and ν and ξ are the control parameters
0 of the enterprise and the investor. It is assumed that when ν=1,
the profit of the enterprise goes to them for investment, and
where λ is the discount rate of financial flows. when ν=0, it goes into its own savings. Similarly, when ξ=1,
To determine the optimal investment process, it is investments are made, and when ξ=0, there is no investment.
necessary to integrate equation (1) taking into account the Additionally, there is the following restriction: ν and ξ must
conditions for G(T). If the investment is constant vary in the range [0, 1]. At the same time, an instant
(D(t)=D0=const), the solution to this problem can be obtained investment is taken, the analysis uses a single-factor linear
in the following form: G0=D0(1-e-λT)/λ. From this expression it production function, and the change in the company's profit
is clear that the total investment costs depend linearly on the relative to fixed assets also has a linear relationship.
rate of investment. Since the linear function has no extremes, The objective function of an enterprise (G1m) is both to
it is not possible to optimize the investment process. To increase its own savings and to increase the value of its fixed
optimize the investment process, instead of linear, we will use assets, and the objective function of an investor (G2) is to get
a quadratic function, and then the investor quality functional maximum profit:
will have the following form:
T
G1m 1 e t dt X T max ,
T
G1 D (t )e 2 t
dt
min .
0
0
Where , based on the obtained expressions μ=0.10 1.1 1.6 2.3 3.2 4.4 5.9
for D*(t) and D0, the total costs of the investor were calculated μ=0.15 1.6 2.1 3.2 4.5 5.7 8.0
using the expression for G(T).
μ=0.20 2.3 3.3 4.5 5.7 7.3 9.6
The calculations have shown that managing the investment
μ=0.25 3.3 4.3 5.6 7.3 8.9 11.1
process with D*(t) allows the investor to reduce his total costs
103 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
After analyzing the task, it was obtained that for In the next stage of the research, optimization of the
integrating equation (3) with due regard for requirements (4), management of the joint venture for the manufacture and sale
Pontryagin's maximum principle can be used, which allows of industrial products is carried out. It discusses the work of
one to obtain optimal trajectories of control parameters ν and ξ the MAS, consisting of several agents that are enterprises of
in the presence of restrictions on them. As a result of solving EGs, for example, auxiliary and main production (Fig. 3). The
the problem, it was obtained that over time of joint work [0, auxiliary production enterprise (agent A) acquires the
T], the control parameters ν and ξ should change as follows: necessary materials and components, conducts their control,
preprocessing and then transfers them to the main production
for manufacture, testing, assembly and further implementation
(agents B and C). The joint work of agents over time in such
MAS can be represented by the following system of
differential equations:
As can be seen, changes in the control parameters are of a x' U X 1 x V Z1 z RY 1 y ;
relay nature and equal to one of their two possible limit z ' V Z1 z Wz ;
values. The moments of time at which changes in these y ' RY 1 y Sy.
parameters should occur are equal to: (5)
b 1 bL Here x , y , and z accordingly, the quantity of products
t0 T ; t1 t 0 ln ;
b bL in the warehouse of agents A, B and C, U(t) is the speed of
1 bL production by agent А, V(t),W(t),R(t),S(t) the speed of the
t 2 T ln rate acquired and subsequently sold by agents B and C units of
bL production, X1, Z1 and Y1 are the maximum production
b / X , capacities of the warehouses of agents A, B and C.
Where , wherein
During the joint work [0,T], each of the agents seeks to get
t1 t0 t 2 T . the maximum profit for themselves (J1 for agent A, J2 and J3
for agents B and C):
In order to fulfill conditions (4), the optimal management
J1 c1 V Z1 z c3 RY1 y c0 U X 1 x d1 xdt max
T
of the use of its profits by an enterprise will be as follows. On
the time interval [0, t0], where ν (t)=0, the profit of the
enterprise is sent to its own savings, and on the time interval 0 (6)
[t0, T], where ν (t) = 1, it is spent on the IP (Fig. 2 ).
J 2 c2Wz c1 V Z1 z d 2 z dt max
T
104 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
where c0, c1, c2, c3, c4 are the unit costs manufactured by
agent A, the products purchased and sold by agents B and C,
and d1, d2 and d3 are the additional costs of transporting and
storing them in the agents' warehouses.
Each agent can get the maximum value of its own
functional quality by changing its control parameters U (t)
(agent A), V (t), W (t) (agent B) and R (t), S (t) (agent C).
Consider first the work of agent B in such MAS. To do
this, it is necessary to integrate (5) taking into account the
fulfillment of the requirements (8). In solving this problem,
we will use the Pontryagin maximum principle, for this we
construct the Hamilton function:
H 3 1 U X 1 x h1 RY1 y h2 Sy h3 V Z1 z d3 y, Fig. 4. Algorithm for Determining the Optimal Parameters for Managing
Agents.
Where a ψ1, ψ2 and ψ3 are auxiliary variables defined
by the expressions: The determination of the optimal modes of operation for
agent A was also carried out using the Pontryagin maximum
H3 H H principle, using the system of equations (5) and (6). As a
1' U 1 ; 2' 3 Vh3 W 2 ; 3' 3 Rh1 Sh2 d 3
x z y result of the analysis, it was determined that in order to obtain
maximum profit, Agent A should produce products in the
In accordance with the Pontryagin maximum principle, the initial period of time [0,t0], and stop production in the period
optimal control of the work of agent C will be if the Hamilton of time [t0,T]. Fig. 4 shows the scheme for determining the
function has the maximum value. This will be the case if the control parameters of agents A, B and C when they work
control parameters of agent C are changed as follows: together in the MAS.
When determining the point in time t0, it is necessary to
take into account that Agent A usually tries to fully realize all
products manufactured for them, i.e. receive x(T ) 0 . In
this stage of the research, an expression is obtained for
It can be seen that the changes in the control parameters of determining the quantity of products in the warehouse of agent
the agent C are of a relay nature and are equal in magnitude to A at the end of the collaboration time:
one of their two possible limiting values. Moreover, the whole
process of changing these parameters consists of two intervals,
xT K 0 K1 x0 K 2 t1 t0 K3 e V W t1 e V W t 0 , (10)
in which these parameters have a constant value, and the
duration of these intervals is determined by the auxiliary Where К0,К1,К2 and К3 are constant values depending on
variables ψ1, ψ2 and ψ3, which can be found from the the initial indicators of the joint work of agents. Equating
solution of the differential equations for them. The boundary expression (10) to zero, we can find the unknown value t0.
values for these variables were defined in this part of the Due to the fact that the resulting equation is non-linear, its
research, and they are respectively equal: solution is carried out by numerical methods. For this purpose,
ψ1(T) = ψ2(T) = ψ3(T) = 0. a program is compiled, the algorithm of which is shown in
Fig. 5. Its input data are the left (a) and right (b) boundaries of
The expressions for auxiliary variables obtained in this the interval in which the desired one (t0) is located, as well as
chapter showed that the function h2>0 on the entire interval
the accuracy of calculations (ξ).
[0, T], the function h1>0 on the interval [0, t2], and on the
interval [t2, T] the function h1<0. The time t2, at which the
control parameter R*(t) becomes zero, and agent C stops
purchasing products from agent A, is equal to:
1 c c S d3
t 2 T ln 4 3 ; t1 T 1 ln c2 c1 W d 2
S c4 S d 3 W c2 W d 2 (9)
The work of Agent B is considered in a similar way. It has
been obtained that the optimal control of the process of his
work will be the same as for Agent C, i.e. at the initial time
interval [0, t1], he acquires, and at the time interval [t1, T]
stops purchasing products from agent A. The value of time t1,
at which its control parameter V*(t) changes, is determined by Fig. 5. Algorithm for Determining the Optimal Moment of Time t0 Stopping
9). Production of the Agent A.
105 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
1 V1 1 w
T1 ln ; T2 ln
V V1 Vxk W w W zk
106 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
107 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[3] Othman, S.B., et al., An agent-based decision support system for [12] Al-Rababah, A. and N. Hani. Component linked based system. in
resources' scheduling in emergency supply chains. Control Engineering Modern Problems of Radio Engineering, Telecommunications and
Practice, 2017. 59: p. 27-43. Computer Science, 2004. Proceedings of the International Conference.
[4] Chan, S.H., et al., Decision support system (DSS) use and decision 2004. IEEE.
performance: DSS motivation and its antecedents. Information & [13] Al-rababah, A.A. and M.A. Al-rababah, Module Management Tool in
Management, 2017. 54(7): p. 934-947. Software Development Organizations 1. 2007.
[5] Goodwin, P. and G. Wright, Decision Analysis for Management [14] 14. Sahir, S.H., R. Rosmawati, and R. Rahim, Fuzzy model tahani as a
Judgment 5th ed. 2014: John Wiley and sons. decision support system for selection computer tablet. Int. J. Eng.
[6] Noe, R.A., et al., Human resource management: Gaining a competitive Technol, 2018. 7(2.9): p. 61-65.
advantage. 2017: McGraw-Hill Education New York, NY. [15] Al-Rababah, A.A., T. AlTamimi, and N. Shalash, A New Model for
[7] Rose, D.C., et al., Involving stakeholders in agricultural decision support Software Engineering Systems Quality Improvement. Research Journal
systems: Improving user-centred design. International Journal of of Applied Sciences, Engineering and Technology, 2014. 7(13): p. 2724-
Agricultural Management, 2018. 6(3-4): p. 80-89. 2728.
[8] Ruiz, P.A.P., B. Kamsu-Foguem, and D. Noyes, Knowledge reuse [16] Caniëls, M.C. and R.J. Bakens, The effects of Project Management
integrating the collaboration from experts in industrial maintenance Information Systems on decision making in a multi project environment.
management. Knowledge-Based Systems, 2013. 50: p. 171-186. International Journal of Project Management, 2012. 30(2): p. 162-175.
[9] de Souza Melaré, A.V., et al., Technologies and decision support [17] Al-Rababah, A.A. and M.A. Al-Rababah, Functional Activity Based
systems to aid solid-waste management: a systematic review. Waste Comparison Study for Neural Network Application. IJCSNS, 2007.
management, 2017. 59: p. 567-584. 7(1): p. 153.
[10] Garousi, V., et al. Industry-academia collaborations in software [18] Al Ofeishat, H.A. and A.A. Al-Rababah, Real-time programming
engineering: An empirical analysis of challenges, patterns and anti- platforms in the mainstream environments. IJCSNS, 2009. 9(1): p. 197.
patterns in research projects. in 21st International Conference on [19] Gupta, P. Accelerating datacenter workloads. in 26th International
Evaluation and Assessment in Software Engineering (EASE 2017). Conference on Field Programmable Logic and Applications (FPL).
2017. ACM. 2016.
[11] Van Huben, G.A. and J.L. Mueller, Data management system for file [20] AlRababah, A., Digital Image Encryption Implementations Based on
and database management. 2000, Google Patents. AES Algorithm. VAWKUM Transactions on Computer Sciences, 2017.
13(1): p. 1-9.
108 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In the current era, website developers recognize optimizing problems with certain limitations while fuzzy logic
usability evaluation as a significant factor in the quality and is the basis of knowledge in fuzzy systems and controls [10].
success of e-commerce websites. Fuzzy Analytical Hierarchy
Process (FAHP) is one method to measure the usability of the Logarithmic Fuzzy Preference Programming (LFPP) is one
website. Several researchers have applied Logarithmic Fuzzy method to evaluate usability based on FAHP method [11].
Preference Programming (LFPP) approach to deriving crisp LFPP is an approach using non-linear programming to derive
weight from fuzzy pairwise comparison matrix of FAHP the weight of criteria. LFPP uses the logarithm of natural
approach. However, there is a lack of LFPP method in numbers to repairing Fuzzy Preference Programming (FPP)
determining the consistency index of the decision-maker approach that caused the negative of fuzzy membership degree
judgment. In some cases, LFPP method will produce a [12], [13]. However, the LFPP method has its drawbacks; in
consistency value of 0 from consistent fuzzy comparison some cases, a pairwise comparison matrix that consistently
matrices. This value indicates there is a contradiction with what produces a value of 0 [14], [15]. This case is not in line with
the previous researchers have said, that a constant matrix value the definition that states that the more consistent fuzzy pairing
should be more than 0. This research proposes the extended comparison matrix, the optimal value (*) is closer to 1 [16].
Consistent Fuzzy Preference Relation (ECFPR) to assist the There is a presumption that the probability of a value of 0 in *
regular judgment for specifying the weights in measuring e- is due to a matrix that is not consistent. Therefore, before
commerce website usability. The CFPR method used to form a
calculating the weights using the LFPP method, it is necessary
new pairwise comparison matrix. ECFPR was calculating the
to ensure the consistency of the model. The technique for
lower and upper values at the fuzzy triangular number from the
only n-1 comparison, where n is the number of criteria. The
guaranteeing a matrix to be consistent is called Consistent
numerical experiment showed that the consistency index Fuzzy Preference Relation (CFPR). CFPR is an approach to
obtained by extended CFPR method was more significantly reduce the number of comparisons that are often done by users
better than LFPP method. It was revealed that the optimal value in determining preferences between criteria or alternatives. The
always more than 0. The consistency index of ECFPR method weakness of CFPR only considers the modal value at triangular
has a higher mean value than LFPP, so that the use of the fuzzy number so that it produces a comparison matrix with
ECFPR method can improve the amount of consistency crisp numbers. The extended CFPR is reviewing the upper and
comparison matrices. The ECFPR method was also successfully lower bound and creating a comparison matrix in pairs with
implemented with the experimental case on evaluating e- fuzzy triangular numbers. By applying the extended CFPR
commerce website usability. method, it is expected to increase fuzzy preference
relationships consistency provided by experts to make it better.
Keywords—Usability; e-commerce; website quality; logarithmic The extended CFPR based on user judgment is expected to be
fuzzy preference programming; consistent fuzzy preference valid and consistent so that it can give good weight also from
relations each usability criteria.
I. INTRODUCTION The organization of this paper is as follows. Section 2
Current usability measurement methods do not yet have the briefly reviews the LFPP and illustrates its consistency equal to
right uniformity and agreement on standards in software [1], 0. Section 3 proposes the extended CFPR to create the fuzzy
[2]. One measure for websites usability is the sum of products pairwise comparison matrix. Section 4 explained the numerical
between the weights of the criteria and the value of each of the case of usability evaluation method using extended CFPR and
criteria [3]–[5]. Researchers often regard the weighting of LFPP. The paper concludes in Section 5.
standards as a multi-criteria decision-making problem, given
II. LITERATURE REVIEW
its complex structure. They usually break down the complex
issues into its elements in a hierarchy. Several researchers have A. Logarithmic Fuzzy Preference Programming
conducted usability measurements using a combination of Wang and Chin (2011) reforming the Fuzzy Preference
fuzzy numbers and Analytical Hierarchy Process (FAHP) [6]– Programming (FPP) weight derivation. They modified the FPP
[9]. Fuzzy numbers consider the uncertainty and doubt factors method by adding natural logarithms function to improve the
in the experts in determining the level of importance between negative membership degree and arising in multiple optimal
criteria. Fuzzy logic and fuzzy decision making are part of the solutions. Negative value to makes the expected solution less
branch of fuzzy theory. In fuzzy decision making consider
*Corresponding author: arisn@ugm.ac.id. This research was supported by the Ph.D.
grant of the Indonesia endowment fund for education (LPDP) ministry of Finance
Republik Indonesia for Tenia Wahyuningrum (No. PRJ-4811/LPDP.3/2016).
109 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
valid [16]–[19]. The fuzzy pairwise comparison matrix from represented by a pairwise matrix comparison P=pij, where the
expert judgment can be expressed as size is n×n, pij=µp(ai ,a j ) , i, j {1,2, , n} and the value
A ( aij ) nn of the membership function of fuzzy logic. Elements in the
pairwise comparison matrix are calculated using several
(1,1,1) (l1n , m1n , u1n ) propositions [20].
A
(l 21 , m21 , u 21 ) (1,1,1) (l 2 n , m2 n , u 2 n ) Proposition 1. Consider set criteria or alternatives,
X={x1 , x2 , , xn } associated with a reciprocal multiplicative
(l , m , u )
n1 n1 n1 (1,1,1) (1) preference relation A=(aij) for aij [1/9, 9]. Then, the
corresponding reciprocal fuzzy preference relation, P=pij with
where n is the number of criteria, lij 1/uji, mij 1/mji, uij pij [0,1] associated with A is given as
1/lji and 0 lij mij uij for all i, j 1, 2, …, n, j≠i. To find a
n 1
crisp priority vector Ww w, …, wn)T 0 with wi for pij g (aij ) (1 log 9 aij )
i 1 2 (5)
the fuzzy pairwise comparison matrix [16]. The approximate
Proposition 2. For each P=g(A), where P=(pij), the booth of
equation uses natural logarithmic numbers for the improvement
equations (6) and (7) are equivalent.
of fuzzy pairwise matrix (1).
The LFPP method formulated as Minimize 3
pij p jk pki , i, j , k
2 n 1 n 2 2
2 (6)
J (1 λ ) M δij ηij
i 1 j i 1
(2) 3
pij p jk pki , i j k
Subject to 2 (7)
mij Proposition 3. For each P=(pij), the booth of equations (8)
xi x j λ ln δij ln lij ,i 1,2,...,n 1;j i 1,...,n,
and (9) are equivalent.
lij
u 3
xi x j λ ln ij ηij ln uij ,i 1,2,...,n 1;j i 1,...,n, pij p jk pki , i j k
m
ij 2 (9)
λ,xi 0,i 1,2,...,n,
δij ,ηij 0,i 1,2,..., n 1;j i 1,...,n,
pi ( i 1) p(i 1)(i 2) p(i k 1)(i k ) p(i k )i k21 , i j.
(10)
where xi ln wi for i 1, 2, …, n and M is a specified large Proposition 3 is used to construct a consistent fuzzy
number such as M . Equation (3) can be used to calculate preference relation from the set of n-1 values
the weight of each criterion : { p12 ,p23, ,pn-1n } . A decision matrix with entries that are not
* in the interval [0,1], but in an interval [-k,1+k], k>0, can be
* exp( xi )
wi n
, i 1,2,..., n, obtained by transforming the result values using a
*
exp( x j ) transformation function that preserves reciprocity and additive
j 1 consistency. It is given by the function f:[-k,1+k] to [0,1],
(3)
f(x)=(x+k)/(1+2k).
* xi*
where exp() is the exponential function exp( xi ) e for i III. PROPOSED METHOD
1, 2, …, n.
A. Extended Consistent Fuzzy Preference Relation
B. Consistent Fuzzy Preference Relation Wang and Chin (2011) argued that the consistency index of
Preference relations are usually constructed as a matrix that fuzzy pairwise comparison can be seen at the value of * and
represents the degree of interest for the first criteria over the *[16]. The inconsistency in a fuzzy pairwise comparison
second criteria. The relationship of this assessment can be matrix could be expressed as a Proposition 4.
multiplicative preference relations of fuzzy preference
relations. Multiplicative preference relations can be formulated Proposition 4. If (* > 0) (* = 0) then the matrix is very
as inconsistent.
Proposition 4 can be broken down into two new schemes p
R A A, R (rij ), i, j {1,2, , n}, and q. p = (* > 0) (* = 0) and q = the matrix is strong
inconsistent. The proposition can be explained as pq, then
where A is the set of criteria or alternatives, rij is the the equivalent of Proposition 4 is qp. It is also can be
preference ratio of criteria or alternative ai to aj, aijaji=1, i, j expressed as a Proposition 5.
{1, 2, …, n}. Furthermore, the relationship will be
110 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
q=
x1 x3 λ ln 1 δ13 ln 1,
q( ≤ 0) ( ≠ 0) = true
* * x1 x3 λ ln 1 13 ln 1,
x1 x4 λ ln 5 4 δ14 ln 2 5,
If r = (* ≤ 0), s = (* ≠ 0) then qr s. Because of the
x1 x4 λ ln 4 3 14 ln 2 3,
n 1 n
ij ij
*
i 1 j i 1
*2 *2
is always more than equal to 0, then x2 x3 λ ln 5 4 δ23 ln 2 5,
the proposition r is always false. For the statement to be true, x2 x3 λ ln 4 3 23 ln 2 3,
the * must be not equal to 0. It can be said, that the x2 x4 λ ln 3 2 δ24 ln 2 3,
consistency of the fuzzy pairwise matrix must be not equal 0
(* ≠ 0). The condition of this proposition to comply with the x2 x4 λ ln 5 4 24 ln 2 5,
truth table of OR operations. x x λ ln 5 4 δ ln 2 5,
3 4 34
construct fuzzy evaluation matrix on ECFPR method. If the
J 1 M ij ij
2 3 4 2 2
strong importance of element j over element i holds, then the
i 1 j i 1
pairwise comparison scale can be represented by the fuzzy
Subject to number where 0 < lij < mij < uij, for all i, j = 1, 2, …, n, j≠i. For
each pij where i=j+1 or i=j, the elements can be transformed as
(12).
111 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
(u , m, l ) ( i k 1)(i k ) (u , m, l ) ( i k ) i k 1 k 1 k 1
, 2 , 2 , I can be written as Minimize
2
J 1 M . ij ij
2 3 4 2 2
i j. (13) i 1 j i 1
112 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
C C1 C2 C3 C4
C1 (0.5, 0.5, 0.5) (0.41, 0.5, 0.59) (0.21, 0.35, 0.5) (0.01, 0.2, 0.41)
C2 (0.41, 0.5, 0.59) (0.5, 0.5, 0.5) (0.3, 0.35, 0.41) (0.1, 0.2, 0.32)
C3 (0.5, 0.65, 0.79) (0.59, 0.65, 0.7) (0.5, 0.5, 0.5) (0.3, 0.35, 0.41)
C4 (0.59, 0.8, 0.99) (0.68, 0.8, 0.9) (0.59, 0.65, 0.7) (0.5, 0.5, 0.5)
113 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
2 3
J 1 M ij2 ij2
comparison ratings dan
of criteria 2
matrices based on recommendation
expert judgment i 1 j i 1
Subject to
Fig. 1. Usability Evaluation using ECFPR Method. x1 x2 λ ln 0.87 0.82 δ12 ln( 0.82),
x x λ ln 0.91 0.87 η ln( 0.91),
Website Usability 1 2 12
x1 x3 λ ln 0.91 0.87 η13 ln( 0.91),
x2 x3 λ ln 0.5 0.5 δ23 ln( 0.5),
C1 C2 C3
Loading time Response time Page Size
12 12 13 13 23 23
Fig. 2. The Hierarchical Model of Website usability Evaluation.
Define M = 101, the result value of *, *, *, and x can be
TABLE. III. MEMBERSHIP FUNCTION LINGUISTIC SCALE [6] represented as follows:
Convert from AHP scale to Fuzzy number Linguistic expressions
12* = 8.16 × 10-10, 12* = 0.2229, x1 = 0.9060,
* = 0.76,
1 = (1,1,1) Equal
2 = (1,2,3) Equal-moderate
13* = 0.2229, 13* = 2.622, x2 = 0.8118, 23
*
= 8.94 × 10-10,
TABLE. IV. EXPERT JUDGMENT COMPARISON MATRIX TABLE. V. ECFPR MATRIX COMPARISON
Criteria C1 C2 C3 Criteria C1 C2 C3
C1 (1, 1, 1) (4, 5, 6) p13 C1 (0.5, 0.5, 0.5) ( 0.82, 0.87, 0.91 ) ( 0.82, 0.87, 0.91 )
C2 p21 (1, 1, 1) (1, 1, 1) C2 ( 0.09, 0.13, 0.18 ) ( 0.5, 0.5, 0.5 ) ( 0.5, 0.5, 0.5 )
C3 p31 p32 (1, 1, 1) C3 ( 0.09 ,0.13, 0.18 ) ( 0.5, 0.5, 0.5 ) ( 0.5, 0.5, 0.5 )
114 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
115 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[6] P. D. D. Dominic dan H. Jati, “A comparison of Asian airlines websites [17] S. Dožić, T. Lutovac, dan M. Kalić, “Fuzzy AHP approach to passenger
quality : using a non-parametric test,” Int. J. Bus. Innov. Res., vol. 5, no. aircraft type selection,” J. Air Transp. Manag., vol. 68, 2017.
5, 2011. [18] M. Balouchi dan E. Khanmohammadi, “Using logarithmic fuzzy
[7] I. Masudin dan T. E. Saputro, “Evaluation of B2C website based on the preference programming to prioritization social media utilization based
usability factors by using fuzzy AHP & hierarchical fuzzy TOPSIS,” in on tourists’ perspective,” Found. Manag., vol. 7, no. 1, hal. 7–18, 2015.
IOP Conference Series: Materials Science and Engineering, 2016, vol. [19] M. Momeni, A. Sasani, M. R. Fathi, dan E. Khanmohammadi, “Applying
114, hal. 1–8. Logarithmic Fuzzy Preference Programming for Ranking of Effective
[8] S. Aydin dan C. Kahraman, “Evaluation of E-commerce website quality Organizational Factors on Creativity : A Case Study Mansour Momeni
using fuzzy multi-criteria decision making approach,” IAENG Int. J. Faculty of Management University of Tehran,” Int. J. Bus. Soc. Sci., vol.
Comput. Sci., vol. 39, no. 1, hal. 64–70, 2012. 3, no. 14, hal. 83–95, 2012.
[9] N. Sehra, Sumeet Kaur Brar, Yadwinder Singh Kaur, “Applications of [20] E. Herrera-Viedma, F. Herrera, F. Chiclana, dan M. Luque, “Some issues
Multi-criteria Decision Making in Software Engineering,” Int. J. Adv. on consistency of fuzzy preference relations,” Eur. J. Oper. Res., vol. 154,
Comput. Sci. Appl., vol. 7, no. 7, 2016. no. 1, hal. 98–109, 2004.
[10] L.-X. Wang, “A Course in Fuzzy Systems and Control.” Prentice Hall, [21] T. L. Saaty, “How to make a decision : The Analytical Hierarchy
Hongkong, hal. 1–441, 1997. Process,” Eur. J. Oper. Res., vol. 48, hal. 9–26, 1990.
[11] R. Wardoyo dan T. Wahyuningrum, “University Website Quality [22] R. W. Saaty, “The analytic hierarchy process-what it is and how it is
Ranking Using Logarithmic Fuzzy Preference Programming,” Int. J. used,” Math. Model., vol. 9, no. 3–5, hal. 161–176, 1987.
Electr. Comput. Eng., vol. 8, no. 5, hal. 3349–3358, 2018. [23] H. Khademi-Zare, M. Zarei, A. Sadeghieh, dan M. Saleh Owlia,
[12] L. Mikhailov, “A fuzzy programming method for deriving priorities in “Ranking the strategic actions of Iran mobile cellular telecommunication
the analytic hierarchy process,” J. Oper. Res. Soc, vol. 51, hal. 341–349, using two models of fuzzy QFD,” Telecomm. Policy, vol. 34, no. 11, hal.
2000. 747–759, 2010.
[13] L. Mikhailov, “A fuzzy approach to deriving priorities from interval [24] R. J. Chao dan Y. H. Chen, “Evaluation of the criteria and effectiveness
pairwise comparison judgements,” Eur. J. Oper. Res., vol. 159, no. 3, hal. of distance e-learning with consistent fuzzy preference relations,” Expert
687–704, 2004. Syst. Appl., vol. 36, no. 7, hal. 10657–10662, 2009.
[14] E. Iryanti dan R. Pandiya, “Application of Logarithmic Fuzzy Preference [25] M. Celik, I. D. Er, dan A. F. Ozok, “Application of fuzzy extended AHP
Programming for Determining Priority as An Institutional Development methodology on shipping registry selection : The case of Turkish
Strategy,” in International Conference on Computer Applications and maritime industry,” Expert Syst. Appl., vol. 36, no. 1, hal. 190–198,
Information Processing Technology (CAIPT), 2017, hal. 1–5. 2009.
[15] Y. M. Wang dan K. S. Chin, “A linear goal programming priority method [26] A. A. Ali dan P. D. D. Dominic, “A Case Study of Linear Weightage
for fuzzy analytic hierarchy process and its applications in new product Model for Supplier Selection Process,” 2008 Int. Symp. Inf. Technol.,
screening,” Int. J. Approx. Reason., vol. 49, no. 2, hal. 451–465, 2008. vol. 3, hal. 23–26, 2008.
[16] Y. M. Wang dan K. S. Chin, “Fuzzy analytic hierarchy process: A [27] T. Tullis dan B. Albert, Measuring The User Experience. USA: Morgan
logarithmic fuzzy preference programming methodology,” Int. J. Approx. Kaufmann, 2013.
Reason., vol. 52, no. 4, hal. 541–553, 2011.
116 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Background of this research is the globalization development of information needs, especially technology in
and development of science, especially in the field of information the globalization era, has had a significant influence on the
and communication technology and communication that has application of information systems in the health sector. Health
influenced and has implications for changes and renewal of Information System is a set of arrangements that includes
people's lives, including in the field of nursing. So that the role of data, information, indicators, procedures, tools, technology,
information and communication in this aspect of life is very and human resources that are interrelated and managed in an
important, even the futurists, for the most part, have an integrated manner that provides information support for
agreement that one of the most important strengths as the source decision making processes, health program planning,
of future power is information. Purpose: identify the use of
implementation of monitoring and evaluation at every level of
information technology in nursing to determine the effectiveness
of the use of information systems in nursing, identify nurses
health administration Health information systems are a very
'knowledge about the effectiveness of nursing information important part of improving hospital efficiency and supporting
systems, identify nurses' knowledge seen from the socialization of competitiveness by providing health service information for
the effectiveness of nursing information systems. Method: management [1].
Quantitative Research Type with a survey approach conducted Nursing services in a hospital environment is one of the
on 220 nurses. Significant validity test is <0.05, Cronbach Alpha services in the health sector that has an important role in
reliability test> 0.60. The data is then tested in a classic determining the success of services provided in hospitals.
assumption test consisting of multicollinearity tests,
With the highest number of nurses in the hospital
autocorrelation tests, heteroscedasticity tests, normality tests,
multiple linear regression, t-tests, F tests, coefficient of
environment, the existence of nursing services must be
determination tests. Results: the use of information technology managed properly to produce the quality of nursing services
affects the effectiveness of nursing information systems. Nurse provided. Improving the quality of nursing information
knowledge does not affect the effectiveness of nursing systems is one solution to improve the quality of nursing
information systems. Nurse knowledge seen from socialization services. It is undeniable that so far the development of
does not affect the effectiveness of nursing information systems. nursing information systems in this country has not been
The use of information technology and nurse knowledge going well. Nursing information systems are a combination of
influences the effectiveness of nursing information systems. The computer science, information science and nursing science
results of the coefficient of determination that affect the use of that are organized to facilitate management and the process of
information technology, knowledge of nurses, socialization as a gathering information and knowledge used to support the
control variable on the effectiveness of nursing information implementation of nursing care. Meanwhile, according to
systems. Suggestion: Hospital managers must pay attention to the America Nursing Asociation [2] nursing information systems
quality of nursing human resources, through training, relating to legality to obtain and use data, information and
certification, recognition of competencies, supervision, selection, knowledge about documentation, communication standards,
and guidance aimed at improving safe, comfortable and support the decision making process, develop and disseminate
satisfying services for patients, families, communities. new knowledge, improve quality, effectiveness and efficiency
of care and empower patients to choose the desired health
Keywords—Information systems; knowledge; nursing;
socialization
care. The reliability of an information system in an
organization lies in the interrelationship between existing
I. BACKGROUND components so that it can be produced and flowed into useful,
accurate, reliable, detailed, fast, relevant information for an
Information systems are computer systems that collect, organization. This information system is expected to improve
store, process, retrieve, show, and communicate information service quality in achieving service quality standards. Clinical
needed in practice, education, administration and. Many indicators of service quality include: measurement of
benefits are obtained in the use of information systems. This decreased patient rates, pressure sores, nosocomial
benefit not only reduces errors and increases speed and pneumonia, nosocomial infections, and incidences of medical
accuracy in care, but also reduces health costs by coordinating errors. This computer-based information system will identify
and improving service quality. The rapid progress and
117 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
various types of patient needs, ranging from care This is a very large capital that strongly supports the
documentation, medication documentation, to financial implementation of nursing information systems.
calculations that must be paid by patients for the care they
have received [3]. Many people complain about the health Now the only problem is how we are committed, from top
services they receive from nurses. management to the lowest management to fight for the
implementation of nursing information systems in each
For this reason, nurse performance needs to be improved nursing service unit. The reason for the lack of availability of
so that the quality of care services can be provided properly. funds to develop nursing information systems is a classic
One measure of whether quality health services are provided reason that should no longer exist. Especially seeing the
to the community is the level of satisfaction for the people importance of nursing information systems to improve the
who receive the services themselves. One of the benefits of quality of nursing services in particular and health services in
implementing a nursing information system in a hospital general. Based on the above problem, this research was
setting is to help nurses document nursing care. Nursing care conducted with the aim of 1) identifying the use of nursing
in meeting the basic needs of patients is provided by nurses in information technology on the effectiveness of nursing
various health care settings using the nursing process. Nurses information systems, 2) identifying nurses 'knowledge about
use nursing information systems with the aim to clearly assess the effectiveness of nursing information systems,
patients, prepare nursing plans, document nursing care, and to 3) identifying nurses' knowledge as seen from the
control the quality of nursing care. Nurses can have an socialization of the effectiveness of nursing information
integrated view of data (for example integration between systems. The use of information technology is the use of
nurses and doctors in patient care plans) [4]. By utilizing technology in the Nursing Nursing Service System Nursing is
nursing information systems nurses can save time to do the the understanding of nurses in the operation of Nursing
recording compared to when done manually recording. In Information System software to provide accurate and timely
addition, data recorded using a nursing information system results in care reports so that they have an increasing impact
will be more secure. The risk of recorded data will be lost is on the Hospital. Socialization is the dissemination of
very small. In contrast to paper-based records, where the information about Nursing Information Systems conducted by
possibility of data loss is very likely to occur. In addition, the hospitals to nurses in operating information systems. The
existence of a nursing information system will also increase effectiveness of the Nursing Information System is the
the effectiveness and efficiency of nursing staff work. operation of the Nursing information system in the hospital
Implementation of nursing information systems in hospitals, with ease of use, accurate results, and timeliness.
which combines computer science, information science, and
nursing science that is designed to facilitate management and II. METHOD
the process of collecting data, information, and knowledge to This research is quantitative, use a survey research, with a
support the implementation of nursing care [5]. Nursing sample of 220 nurses from several hospitals in Medan. The
information systems are being developed on an ongoing basis instrument used was tested for validity and reliability, the
in the future nursing knowledge will depend on the ability of results of the analysis showed all items were valid because
information systems to facilitate the results of diagnosis, each indicator showed a significant result that was <0.05. The
management, research, education, information exchange, and reliability test results showed that all research instruments
collaboration / collaboration, that the integration of nursing, were reliable because each instrument produced a Cronbach
computer science can be used to identify, collect, process, Alpha >0.60. The data is then tested in a classic assumption
organize data and information to support nursing practice, test consisting of multicollinearity tests, autocorrelation tests,
administration, education, research, and development of heteroscedasticity tests, normality tests, multiple linear
nursing science [5]. The need for a management information regression, t tests, F tests, coefficient of determination tests.
system supports nurses in assisting decision making.
Advances in technology in hospitals enable nurses to use The hypothesis in this study was formulated as follows:
management information systems to support the delivery of H1: The use of information technology has a significant
nursing care, so that better quality nursing care is achieved [6]. positive effect on the effectiveness of nursing information
According to [7] the research focused on exploring systems.
Computerized Provider Order Entry (CPOE) and its impact on H2: Nurse knowledge has no significant effect on the
the work done by nurses. The result is that CPOE is a effectiveness of nursing information systems
technology designed to replace paper entry, communication
and coordination with automated methods, one of which is in H3: Nurse knowledge seen from socialization has no
collaborative collaboration to prescribe drugs in acute care. significant effect on the effectiveness of nursing information
CPOE has been proven to improve communication efficiency systems.
and reduce drug transcription errors and reduce patient care
time, so the patient's morbidity and mortality are reduced. III. RESULT AND DISCUSSION
Actually to implement a nursing information system in a A. Descriptive Analysis
hospital environment is not too difficult to implement, only a
Descriptive analysis shown in Table I, results about the use
commitment to implement it is needed. In times of technology
of information technology on average showed 4.39 where the
such as now, maybe almost all nurses can operate a computer
average respondent answered agree, the nurse's knowledge
as a device in the application of nursing information systems.
showed an average value of 4.55 where the respondent
118 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
responded most agreed, socialization showed an average value Based on these results it can be estimated that multiple
of 1.9 where the average respondent disagrees. participate in linear regression is as follows: Y = 11,852 + 1,319 X1 + 0.038
the socialization of Nursing Information Systems. The X2 + 0.134X3
effectiveness of the nursing information system shows an
average value of 4.18 where the nurse answers the average in Interpretations of the regression equation are:
the agreement. a: 11,852 means that the use of information technology
(X1), knowledge of nurses (X2) and socialization (X3), is
B. The Classic Assumption Test
equal to zero, so the utilization of information systems (Y) is
In Table II, test results show that there is no classic equal to 11,852.
consideration problem. In the normality test shows the value
of sig. equal to 0.715 > 0.05 which means that the data are b1: 1,319 which means that the influence of nurse
declared normally distributed. Multicollinearity values knowledge (X2) and socialization (X3) variables on the
indicate if tolerance values >0.1 and VIF <10, which means information system capability (Y) is positive, which supports
there are no multicollinearity problems. The autocorrelation the ease of information technology increased by one (unit),
test results showed a probability value of 0.818 > 0.05. can improve nursing information systems (Y) by 1,319 with
Heteroscedasticity test results >0.05 which shows no nurses' knowledge assumptions (X2), and socialization (X3)
heteroscedasticity problems. are considered permanent.
C. Multiple Linear Regression Analysis b2: 0.038 means that the influence of the use of
information technology (X1), socialization (X3) on the ability
This analysis is used to examine the effect of utilizing of a positive accounting information system (Y), which
information technology, nurse knowledge, and socialization. supports nurses' knowledge to increase one (unit), can
Based on data processing using the SPSS program, the improve nursing information systems (Y) by 0.038 with the
following results are obtained as shown in Table III. assumption information technology (X1) and socialization
(X3) are considered permanent.
TABLE. I. DESCRIPTIVE ANALYSIS
b3: 0.134 means that the use of information technology
Variabel Min Max Mean Std. Deviation
(X1), nurse knowledge (X2) for accounting information
Utilization of Information systems (Y) is positive, which supports the socialization of an
4 5 4,39 0,32
Technology increase in one (unit), so as to improve nursing information
Knowledge of Nurses 1 5 5 0,49 systems (Y) by 0.134 assuming benefits information
Socialization 0 12 1,9 2,82 technology (X1) and nurse knowledge (X2) are considered
permanent.
Nursing Information
3 5 4,18 0,37
Systems D. Coefficient of Determination (Adjusted R Square)
Based on the SPSS calculation results, the adjusted R
TABLE. II. CLASSICAL ASSUMPTION TEST square value is 0.324. This shows 32.40 percent, the variable
Multicollinearity
utilization of information technology, knowledge of nurses,
Heteroske- and socialization of the effectiveness of nursing information
Variable Normality
Tolerance VIF dasticity
systems. While the rest (100% -32.4%) = 67.6% is explained
Utilization of by other variables not included in this regression model or not
Information 0,569 1,757 0,757 0,715 examined in this study.
Technology
E. F Test
Knowledge of
Nurses
0,491 2,035 0,373 Based on the calculation of multiple linear regression
shows that Sig. F = 0.026 <α 0.05. This reflects if the model
Socialization 0,822 1,217 0,527 used in this study is suitable. The use of information
technology, nurse knowledge and socialization simultaneously
TABLE. III. MULTIPLE LINEAR REGRESSION TEST
has a significant influence on the effectiveness of nursing
information systems.
Model β t Sig
F. Utilization of Information Technology about the
a costant 11,852 0,682 0,505 Effectiveness of Nursing Information Systems
Utilization of Information The results of the probability value of 0.019 < 0.05 then
1,319 0,504 0,19
Technology,
H1 is rejected as a positive information technology that
Knowledge of Nurses, 0,038 0,64 0,950 supports the use of nursing information systems. The results of
Socialization 0,134 0,282 0,782 this study are supported by research conducted by [8] in which
the results obtained are related to positive technology for the
Adjusted R2 0,324
use of information systems. Thus, the higher the level of
F count 4,042 information technology usage, the higher the level of nursing
Sig. F 0,026 information system requirements. Therefore, the results of this
study support the results of research conducted by [8] and [9].
119 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
120 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[10] Hamza. 2016. Design and Build a Nursing Care Information System for [13] Ningsih, Ratna. 2010. Application of Nursing Information Systems in
Patients with Pneumonia. Yogyakarta: Journal of Information Systems Complete Nursing Documentation at Hospitals. Jakarta
(JSI), VOL. 8, NO. 1 [14] Komala, 2012, The Influence of Knowledge of Accounting Managers
[11] Gomez-Mejia L.R, Balkin, D.B. & Cardy, R.L, 2011, Managing Human and Top Management Support on Accounting Information Systems and
Resources, International Edition. Prentice Hall International, Inc. Their Effects on Information Quality: Survey at the Zakat Management
[12] Ratnaningsih, 2013, The Influence of Information Technology Institute in Bandung. Third International Conference on Business and
Sophistication, Management Participation, Management Participation Economic Research (Third Celebration 2012).
Knowledge, and Accounting Manager Knowledge on the Effectiveness [15] Soudani S N, 2012, The Use of Information Systems for Effective
of Information Systems in Starred Hotels in Badung Regency, Faculty of Organizational Performance. International Journal of Economics and
Economics, Udayana University. Finance, 4 (5), Pp: 136-145.
121 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Real-time motion detection in specific area is Before 2008, the theoretical test was in form of
considered the most important task in every video surveillance Asking/answering test between the supervisor and the
system. In this paper, a novel real time motion detection candidate by using a pre-defined list of questions related to
algorithm introduced to process Line zones called Line road situations, prerequisites and notions in order to evaluate
Monitoring Algorithm (LMA). This algorithm integrates the candidate‟s knowledge by calculating the number of correct
Bresenham’s Algorithm and Structural Similarity Index (SSI) in answers, which must exceed the pass- average related to the
order to achieve the best performance. Bresenham’s Algorithm is chosen category. The pass average is specified by the law for
used to collect line pixels from two given points. Then, the SSI is each category (Ex 32/40 for Category “B”: Automobile) when
used for real-time calculation of similarity for line motion
a candidate fails in the exam, he/she will be re-called for the
detection. The most attractive side of using the LMA is that the
algorithm does not need to compare all pixels of the whole images
second session after fifteen days for his/her last chance.
or regions for line areas monitoring. This algorithm has high In 2008, according to the law, the Moroccan government
capability, treatment speed and efficiency for motion detection has decided to employ technology in the theoretical test due to
and also demands less compilation time for the hardware various problems. These problems are caused by kipping
performance. The main objective of this paper is to use a video completely the judgment on supervisors hand besides the time
surveillance system implementing LMA to supervise the Car and resources wasting with not much reliable results.
Reverse Test (CRT) for driving license exam in Morocco. The
Therefore, the proposed solution was to regroup candidates
evaluation of the experiment results in implementing the
into sessions, each one passes the exam in front of a monitor
proposed algorithm is reported in this paper.
related to remotes for each candidate. In so doing, it allowed
Keywords—Bresenham’s Algorithm; Structural Similarity them to answer separately to questions shown in the monitor
Index; SSI; motion detection; Line Monitoring Algorithm; LMA; and get the results by the end of the session. This arrangement
OpenCV; surveillance; camera; video surveillance system made by the government gives more reliable results and solves
a lot of problems. However, there were other problems caused
I. INTRODUCTION by the system such as remotes crashes problems and unsaved
According to the World Health Organization's (WHO) results. Therefore, in 2010, computers having a screen-touch
2015 [1] Road Safety Situation Report, nearly 1.25 million and cameras are used for the first time instead of the old
people die every year on the roads. In Morocco although the system. Thus, each candidate can take anytime his exam
total number of registered vehicles (per capita) is low, the rate separately with random questions. This would give credibility,
of road traffic deaths is high with approximately 10 people transparency and efficiency to the test and evaluate objectively
killed every day. 36% of the fatalities are car occupants and a the candidate‟s abilities and prerequisites and moreover, to
further 21% are motorcycle riders. facilitate the job for the supervisor to become just an observer
instead of being a judge with no pressure.
Road Safety has become a matter of national security in
Morocco. Many arrangements have been made by the On the other hand, practice test did not change. That is,
government to decrease the number of car accidents such as according to Moroccan law [2] the practice test divided on four
installing Speed Radars, Road Signs and Speed Measuring stages for category “B“ correspond to car driving :
Panels without neglecting the idea of improving driving license Parking Car Test: The candidate must successfully park the
exam. vehicle in and out of a specific zone (Figure 1) without
The driving license exam in Morocco is composed of two touching any bars surrounding that zone.
sections: Entering Car to the garage Test: The candidate must
Theoretical test: a sort of quiz which, according to the successfully enter the vehicle in and out of a dedicated vertical
category desired, the candidate must answer correctly to zone (Figure 2) without touching any bars surrounding the
a group questions (Ex: 32/40 for Cars category “B”). zone which simulates the garage.
122 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
123 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
( ) (2)
124 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
125 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
126 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
127 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
measures discussed by the authors of [16] will be adopted. TABLE. II. A CONFUSION MATRIX FOR SYSTEM RESULT CLASSIFICATION
These measures are introduced in this work as the following:
System result/Test Line crossed Line not crossed
-True Line Crossed (TLC): It is the number of every couple
Change declared TLC FLC
of successive frames where the vehicle crosses the line
meanwhile the CRT system declares the change. Change not declared FLN TLN
-False Line Crossed (FLC): It is the number of every
couple of successive frames where the vehicle crosses the line TABLE. III. RESULT OF APPLYING LMA ON TEST VIDEO
but no change is declared by CRT system.
Th 0.95 0.85 0.75 0.45 0.3
-True Line Not Crossed (TLN): It is the number of every
TLN 351 401 422 472 586
couple of successive frames where the line is not crossed while
the CRT system declares the change. TLC 145 95 74 64 57
Specificity 1 1 1 1 1
128 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
As regards to the experimental results from above, it is In the Figure 17, Figure 18 and Figure 19, the proposed
clear that choosing the right value of threshold can be affected form represents the real scene view of CRT system through the
by the quality of the video. In our case, the change in selected use of captured images from the real scene where all
line should be relatively high with the highest value of parameters are drawn. In this stage, we tried to manage the
accuracy and low count of false detected frames FLC. monitoring of Start Line and End Line. In further researches,
Therefore, we conclude that the most accurate threshold value the focus will be on other techniques to process the Tracking
should not exceed 0.3 in order to achieve best results. Area and Limit Lines; these techniques would provide useful
data such as calculating the distance between the vehicle and
C. The Result of the Proposed Solution the limit lines, measuring the vehicle speed, which can be used
In this section, we will use the same video as the subsection to make decisions and helping the learning process of the
4.2 above with a threshold value Th=0.3. The results displayed program.
in Table V show that when using the chosen Th value, CRT
System detects successfully the changes in start line and end TABLE. V. EXPERIMENTAL RESULT OF IMPLEMENTATION OF LMA IN
line with an accuracy of 97%. The histogram represented in CRT
Figure 16 shows the variation of SSI score by the video frames
Scene Accuracy Snapshot
where TLN, TLC and FLC values are highlighted. As regards
the car crossing event, the graph above can be divided into
three parts:
In the first part (1), the SSI score is greater than 0.3 so the Start line 97 %
system did not declare the change and the car is not arrived yet
to the line (TLN). The highlighted part (2) represents the
entering/leaving events of the car where the system declares
the change (TLC). In the next part (3), the SSI score is under
0.3 in several frames which leads the system to declare false
End line 97%
changes (FLC) and that affects the accuracy (the other missing
3%). It is due to the position of the camera, which shows the
line been drown upon the vehicle. Thus, to manage this
problem, the CRT system should not monitor the line when a
change has already detected for the first time.
129 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
130 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[10] Wang, Z., Bovik, A.C., Sheikh, H.R. and Simoncelli, E.P. (2004) Image [14] Z. Zhang, T. Jing, J. Han, Y. Xu, and X. Li, “Flow-Process Foreground
Quality Assessment: From Error Visibility to Structural Similarity. IEEE Region of Interest Detection Method for Video Codecs,” IEEE Access,
Transactions on Image Processing, 13, p.p 600-612. vol. 5, pp. 16263–16276, 2017.
[11] I. Joy Kenneth Breshenham's algorithm. In Visualization and Graphics [15] P. Banerjee, S. Sengupta, "Human motion detection and tracking for
Research Group, Department of Computer Science University of video surveillance", National Conference for Communication, 2008.
California, Davis (1999). [16] M. Solokova, N. Japkowicz, S. Japkowicz, Beyond accuracy, F-score
[12] Youtube video intituled: DOTS Parking Cameras Live Stream by DOTS and ROC : a family of discriminant measures for performance
Parking Cameras diffused live on Feb. 10th 2016 [Online].Available: evaluation , Australian Conference on Artificial Intelligence, vol. 4304,
https://www.youtube.com/watch?v=FuvhI9Vmek0/. [Accessed 15 10 pp. 1015-1021, 2006.
2018].
[13] A. Buades, B. Coll, J. Morel, "Denoising image sequences does not
require motion estimation", Proc. IEEE Conf. Adv. Video Signal Based
Surveil. AVSS Palma de Mallorca Spain, pp. 70-74, September 2005.
131 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstracts—To survive in the fierce competition of customer experience, positive effect from word of mouth
telecommunication industry and to retain the existing loyal effects and excellence customer experience [4].
customers, prediction of potential churn customer has become a
crucial task for practitioners and academicians through The income can be increased by improving service quality
predictive modeling techniques. The identification of loyal and customer satisfaction, redress the customer problems and
customers can be done through efficient predictive models. By timely tackle of complaints registered by customer [5]. The
allocation of dedicated resources to the retention of these process can be more effective by introduction of automation
customers would control the flow of dissatisfied consumers through CRM system. The business capabilities can be
thinking to leave the company. This paper proposes artificial improved by better and integrated knowledge in real time
neural network approach for prediction of customers intending from CRM system and help significantly through better
to switch over to other operators. This model works on multiple decision-making. By extracting detailed tables and report
attributes like demographic data, billing information and usage without spending much time which can better improve
patterns from telecom companies data set. In contrast with other working productivity [6]. Excellent and comprehensive
prediction techniques, the results from Artificial Neural analysis can be executed through provisioning of customer
Networks (ANN) based approach can predict the telecom churn service so individual planning and decision making can
with accuracy of 79% in Pakistan. The results from artificial improved through better learning opportunity. The customer
neural network are clearly indicating the churn factors, hence needs, expectations and behavior can be predicted through
necessary steps can be taken to eliminate the reasons of churn.
information derive from CRM [7].
Keywords—Neural Network; ANN; prediction; churn The prime factors behind customer retention can be
management extracted in order to develop the profitable, loyal and long
lasting relationship with their clients. Without running
I. INTRODUCTION
retention compaign, the telecom operators are consistantly
To survive in dynamic, high service demanding sector of losing significant numbers of customers that is 20% to 40%
telecommunication and to achieve operational excellence, it is each years [8]. By bringing improvement in retention
indispensable to maintain up to date customer relationship phenomena the aount £128 million rang from 20% to 10 %
management system (CRM). This system plays pivotal role in can anually be saved by one of british well know telecom
developing customer satisfaction, loyalty and main interface to operator Orange [9]. Many statistical and datamining
interact with our clients [1]. The CRM system could techniques are introduced to investigate the customer churn
revolutionize the entire system by providing real time prediction. In contrast to market survey, data mining
provisioning of information, improvement in sale process and, techniques are analyzing the information obtained from both
enhancement of customer loyalty, advertisements and increase historical & current data in order to predict the patterns on
the effectiveness of marketing tools [2]. This term is basically historical data and future customer attitudes [10].
emerge from relationship marketing defined as – the type of
marketing where a lot of emphasis is laid to improve the The most common techniques used for prediction are
qualitative, strategic and supportive relationship with already Decision Tree (DT), Logitic Regression (LR), Support Vector
working customers and devise strategy to attract new Machine (SVM) and Neural Network (NN). Further more, the
customers.[3]. In order to survive in extreme competitive decision tree is used for resolution of classificaiton problems
market of telecom, the firms must have to adapt with external to divide the instances into two or more than two classes.
changes, so CRM is proved to be an important monitoring Similarly, logistics regression gives the probability by
tools to detect environmental changes on business operation providing input/oupt fields, set of equation and factors causing
and take tough business decision. CRM system is providing customer churn [11].
both tangible and non-tangible benefits hence former are Most of the companies in today‟s world are suffered badly
providing increasing productivity, increase retention rates and due to switching over of dissatisfied customers famously
profitability, decrease the marketing cost and faster turnaround know as customer churn and departure is mostly done to new
time while later benefit is responsible for customer loyalty and competitor. The acquisition of new customer costs new
satisfaction, take the benefits of segmentation, increase
132 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
company 6 to 7 times more than retaining the existing extreme competition. This activate the CRM department to
customer hence cause lot of profit lose [12]. The most convince churner customer by redressing their issues, offering
probable reason behind the departure of customer is achieving extra service and discount in monthly bills so loss could be
of cheaper offer from another company, expression of prevented. The six steps initiating from data collection to
dissatisfaction from existing operator or successful marketing devising of churn policy [16] is depicted in Fig. 2.
strategy of new company [13].
Predictive Analytics can divided into classifications and
prediction which can further be sub divided into decision tree,
logistic regression, neural network and support vector machine
given in Fig. 1. To protect the profit, brand name and assets of
exiting operator, it is deemed necessary to retain the
customers. Hence collection of these customer data and
necessary prediction would help identification of these
dissatisfied customer and would help utilization of resources
for these targeted customers.
133 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
134 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 7. The Pie Chart (Gender, Education Level, Marital Status and Age).
135 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Descriptive Statistics
N Mean Std. Deviation Skewness Kurtosis
Statistics Statistics Std. Error Statistics Std. Error Statistics Std. Error Statistics Std. Error
Age 20468 45.33 .137 19.625 0.008 .008 .017 -1.199 .034
Education 20468 1.75 .006 0.887 .933 .933 .017 -.101 .034
Gender 20468 0.49 .003 0.500 .47 .047 .017 -1.998 .034
Marital Status 20468 0.49 0.500 .041 .041 .017 -1.998 .034
Valid N(listwise) 20468
TABLE. II. THE CORRELATION AMONG AGE, GENDER, TENURE AND MONTHLY CHARGES
Correlations
age gender tenure MonthlyCharges
Pearson Correlation 1 .661** .623** .791**
age Sig.(2-tailed) .000 .000 .000
N 20468 20468 20468 20468
Pearson Correlation .714** 1 .685** .822**
gender Sig.(2-tailed) .000 .000 .000
N 20468 20468 20468 20468
Pearson Correlation .565** .598** 1 .656**
tenure Sig.(2-tailed) .000 .000 .000
N 20468 20468 20468 20468
Pearson Correlation .786** .822** .756** 1
MonthlyCharges Sig.(2-tailed) .000 .000 .000
N 20468 20468 20468 20468
** Correlation is significant at the 0.01 level (2-tailed).
136 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Similarly, the annealing algorithm boosts the interval center TABLE. III. THE CASE PROCESSING SUMMARY
a0 and a0 –a and a0+a. So the initial setting of all parameters
Case Processing Summary
are:
N Percent
Initial lambda =0.0000005, Initial sigma=0.00005 & Interval
Sample Training 16421 80.20%
center =0 and Interval offset =0.5
Testing 4047 19.80%
The SPSS prefer to use cross-entropy error function rather
than squared error function whenever softmax activation Valid 20468 100.00%
function is selected and applied on output layer. The Excluded 0
mathematical form of cross entropy error function is as below: Total 20468
n
Error =-∑ i=1TjInOj (6)
where n= total number of output nodes
Tj=target value (output node j)
Oj= Actual output value (output node j)
V. RESULTS
The main purpose of this research the generation of
multilayer perceptron neural network (MLP-NN) model that
can actually predict the telecommunication churn by
processing of data obtained from telecom industry according
to factors. Below mentioned Table III details the information
of dataset applied for building ANN model.
In the Table IV provide the total number of neurons used
in input, hidden & output layer and all predictor variables
(income per month, drop call rate, failure call rate, education,
monthly bill amounts, complaint lodges for resolution, unpaid
numbers, call duration- total, call duration- average, internet
subscribers, Voice service used, total calling, occupation and
age).
Moreover, the number of nodes calculated in the hidden
layer is 3 while in the output layer is 2, the activation function
taken for hidden layer is selected as hyperbolic tangent (tanH)
and softmax function in the output layer and error function
used as cross-entropy of network information table.
In network diagram extracted from SPSS results, the
telecom customer churn (Churn=No, Churn=Yes) from inputs
tenure.
Total charge. Similarly, there are three nodes in the hidden
layer and two output nodes determined as churn, No-churn
depend on results.
The network diagram of artificial neural network is
furnished in Fig. 8 which are comprised of input layer, hidden
layer and output layer after putting the data in software. The
parameter estimates table is given in Table V while model
summary in Table VI provide details information about
training and testing data set. For both training and testing
while the cross entropy error is brought down to minimum
possible level during training processing. The power of the
model to forecast the outcome of the model is calculated
through small value of cross entropy error is 1623.861. The
percentage of incorrect prediction is 22.7%. There are 10
consecutive steps completed was performed where there is no
more reduction in error function during testing of sample is
performed categorical dependent variable outcome.
Fig. 8. The Network Diagram of ANN.
137 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Network Information
1 monthly income
2 Call drop rate
3 Call failure rate
4 education
5 Monthly billed amount
6 number of complaints
7 Number of month unpaid
Covariates
8 Total call duration
Input layer
9 Avg. call duration
10 Uses internet service
11 uses voice service
12 Total mins used in last month
13 occupation
14 age
Number of Units
Rescaling Method of Covariates Standardized
Number of Hidden Layers
Hidden Layers(s) Number of Units in Hidden Layer1
Activation Hyperbolic tangent
Dependent Variables 1 Customer churn
Number of Units
Output Layer
Activation Function Softmax
Error Function Cross-Entropy
a. Excluding the bias unit
Parameter Estimates
Predicted
Predictor Hidden Layer1 Output Layer
[churn
H(1:1) H(1:2) H(1:3) H(1:4) H(1:5) H(1:6) H(1:7) [churn =1]
=0]
(Bias) -.202 -.046 -.559 .401 .151 .051 .245
monthly income .362 .130 -.457 .036 -.275 .370 -.278
calldroprate .166 -.201 -.436 -.311 -.361 .459 -.263
callfailurerate .313 -.164 -.038 .125 -.467 .200 .311
education .353 -.426 .106 .384 .376 -.441 .454
monthlybilledamount -.308 .349 .020 .065 -.394 -.002 -.372
Number of complaints .249 .135 .044 -.050 -.017 -.008 -.247
Input
numberofmonthunpaid .205 .110 -.203 -.426 -.180 .345 .368
Layer
Total call duration .236 .473 .279 .328 -.195 -.032 .364
avgcallduration .255 -.037 -.117 .172 -.052 -.058 .262
usesinternetservice .331 -.100 -.439 -.487 .278 .379 -.220
uses voice service -.295 .492 .439 -.119 .002 .337 .363
totalminsusedinlastmonth -.027 .488 -438 -.056 -.027 -.423 .443
occupation .280 -.336 -.194 .075 -.406 .215 -.005
age -.127 -.057 -.243 .164 .366 .355 .249
(Bias) .799 -1.132
H(1:1) -.274 .258
H(1:2) .073 .183
Hidden H(1:3) -.291 .222
Layer1 H(1:4) .086 -.490
H(1:5) .238 -.138
H(1:6) .367 -.180
H(1:7) .007 -.354
138 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Model Summary
For, classification (confusion matrix) is drawn in Similarly, the third box plot actually relates to Yes
Table VII. When the predicted outcome is greater than 0.5 the category although predicted in Category churn=No and fourth
outcome is determined as Yes (churn=Yes). The over plot indicated probability and actual are classified as customer
percentage of training, 77.3% classification of training data churn=Yes.
was performed. To determine the possible cutoff point, the sensitivity
The pseudo-probability is calculated in box plot are drawn versus specificity is classified through receiver operating
in Fig. 9. The dependent variable (Churn=Yes or No), the characteristic curve (ROC) is illustrated in Fig. 10.
pseudo-probabilities obtained from whole dataset is displayed The combine sensitivity and specificity (1-false positive
in box plots and each value greater than 0.5 display correct rate) is showing he random diagonal line is drawn from lower
predictions. The first box plot extended from left to right are left side to upper right side at 45 degree and greater the
customer No churn in the category of No churn, the second accuracy can be achieved depend on the distance from 45
plot indicates the classification of churn churn=No, although degree base line through classification process.
the values related are Yes category.
The mathematical determination of the area under the
TABLE. VII. THE CLASSIFICATION curve can determine in Table VIII. The probability predicted
as 79% displays in the model having the customer churn= Yes
Classification and churn=No are randomly selected clearly reveal the
Predicted pseudo-probability of churn prediction in the category
Sample Observed Percent churn=Yes.
No Yes
Correct
No 14926 0 100.00%
Training Yes 1495 0 0.00%
Overall Percent 100.00% 0.00% 90.90%
No 3679 0 100.00%
Testing Yes 368 0 0.00%
Overall Percent 100.90% 0.00% 90.90%
Dependent Variable: Customer Churn
Area
No 0.79
Churn
Yes 0.79
Fig. 9. The Box Plot.
139 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The cumulative gain that displays the classification of This figure indicate the sensitivity of model according to
telecom customer churn calculated through artificial neural change of each input variable. The greatest impact from
network against classification of prediction through chance. independent variable that is tenure/subscription of customer
The fifth point in the curve at category churn=No (50% and with company classify the customers to either churn =Yes or
40%), meaning if dataset is rated and churn cases are Churn= No. On second position, the churn is mostly affected
predicted through pseudo-probability of category churn=No by total charges.
,then it is not difficult to determine that top 40% cases contain
50% of total cases actually take churn=No. Table IX is giving the importance of each variable used in
main data set. The importance along with percentage is given
Simply, the gain given in Fig. 11 is the determination of in more detail for each variable where age has got the highest
effectiveness of classification model that the correct prediction values equal to 100% followed by call duration (Total) equal
out of total model against the prediction determine without to 63%, complaint lodge 33.3%, call duration(Avg) 25%,
using a model. The greater the distance of baseline curve main education 24%, income per month 23.30%, Drop call rate
incline line the greater the gain a model have, which measure 21.9%, monthly income 20.1%, unpaid numbers 19.1% and
the excellent performance. occupation 11.4% at lowest level. All these percentage values
are illustrated in bar chart provided in Fig. 13 for better
On other hand, lift curve drawn in Fig. 12 also evaluate the understanding and arrange in descending orders for easy
performance of the model according to the portion of understanding.
population and give clear view of benefit of using the model
to scenario where there is no predictive modeling. By TABLE. IX. THE INDEPENDENT VARIABLE IMPORTANCE
comparing the gain curve with lift curve, it is determine that
93% value of gain curve , the lift factor is determine as 2.5 on
Independent Variable Importance
lift curve.
Normalized
Importance
Importance
140 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
141 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[15] Mestre, Maria Rosario, and Pedro Vitoria. "Tracking of consumer [20] Nath, Shyam V., and Ravi S. Behara. "Customer churn analysis in the
behaviour in e-commerce." In Proceedings of the 16th International wireless industry: A data mining approach." In Proceedings-annual
Conference on Information Fusion, pp. 1214-1221. IEEE, 2013. meeting of the decision sciences institute, vol. 561,. (2013) pp. 505-510.
[16] V. Umayaparvathi, K. Iyakutti,, “Attribute Selection and Customer [21] Liu, Peng, Naijun Wu, Chuanchang Huang, Bingrong Wei, Libo Wang,
Churn Prediction in Telecom Industry”, Proceedings of the IEEE and Zhen‟an He. "Research on Applying Data Mining to Telecom
International Conference On Data Mining and Advanced CRM." In International Forum of Information System Frontiers-2016
Computing,(2019). Xian International Symposium.
[17] Chen, Kuanchin, Ya-Han Hu, and Yi-Cheng Hsieh. "Predicting [22] Jayne, Chrisina, Andreas Lanitis, and Chris Christodoulou. "Neural
customer churn from valuable B2B customers in the logistics industry a network methods for one-to-many multi-valued mapping
case study." Information Systems and e-Business Management 13, no. 3 problems." Neural Computing and Applications 20, no. 6 (2016): 775-
(2015): 475-494. 785.
[18] Arsanl, Taner, and Safa Çimenli. "Churn Analysis and [23] Adebayo, adelaja oluwaseun, and mani shanker chaubey. "data mining
Prediction." International Journal of Computer Science and Information classification techniques on the analysis of student‟s
Security 14, no. 8 (2016): 550. performance." gsj 7, no. 4 (2019).
[19] Junxiang Lu, "Predicting Customer Churn in the Telecommunications [24] Andreu, David, and Marc Torrent. "Prediction of bioactive peptides
Industry -- An Application of Survival Analysis Modeling Using SAS", using artificial neural networks." In Artificial Neural Networks,
In SAS Proceedings, SUGI27, pages 114-127, 2012. Springer, New York, (NY, 2015) pp. 101-118.
142 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Modified seam carving by switching from the Seam carving is supported by Adobe3 , GIMP4 , digiKam5
conventional method to resizing method depending on the object and Image Magick 6 already [2]. Video carving allows
size is proposed. When the object size is dominant in the scene of targeting to content rich time periods in the video stream so
interest, the conventional seam carving shows deformation of that video content is shortened. Meanwhile, some portions of
components in the object. To avoid the situation, resizing method time series of video contents are deleted so that not natural
is applied rather than the conventional seam carving in the object movement appears some time.
proposed method. Also, the method for video data compression
based on the seam carving not only in image space domain but Video carving method proposed here is based on the
also in time domain is proposed. It is specific feature that original distance between histograms of the objects in concern in the
quality of video picture can be displayed when it is replayed. two adjacent frames, the current frame and the next frame. If
Using frame to frame similarity defined with histograms distance the distance is not less than a prior determined threshold, then
between the neighboring frames, frames which have great such frame can be deleted for video data compression. Also,
similarity can be carved results in data is compressed in time the deleted frame number is stored in the header of the video
domain. Moreover, such carved frame can be recorded in the content so that object movement is much natural in playing the
frame header so that the carved frame can be recovered in compressed video contents by referring the deleted frames (the
reproducing the compressed video. Thus, video quality can be deleted frame is replaced to the previous frame) in comparison
maintained, no degradation of video quality at all. Compression
to the conventional video carving method [13]. On the other
ratio is assessed with the several video data. It is obvious that
hand, improved seam carving by switching from the
data compression ratio of the proposed space and time domain
seam carving is greater than that of the conventional space
conventional method to resizing method depending on the
domain seam carving. object size is proposed. When the object size is dominant, the
conventional seam carving shows deformation of components
Keywords—Seam carving; data compression in time and space in the object. Sometimes it would be fanny shapes of the
domains; video data compression objects. To avoid the situation, image size change method is
applied rather than the conventional seam carving in the
I. INTRODUCTION proposed method.
Image data compression methods which allow The method proposed here is preprocessing of the object
compression using removing content less portion of image size detection before applying space domain of seam carving.
from original images (Seam Carving1) are proposed [1]- [8]. If the object size is dominant in the whole image, object shape
Seam carving with OpenCV, Python, and scikit-image 2 is cannot be maintained its shape is the conventional seam
available. The graph cut concept was proposed together with carving applied to the image of concern. To avoid the
dynamic graph cut based on Markov random fields [7]. These situation, object size is detected prior to the seam carving in
methods allow image segmentation, texture extraction, image the proposed method. Also, it is possible to compress the
mosaic, image energy minimization, etc. Also, video cutout video data in concern by considering the object moving speed.
method is proposed for targeting an image portion through If the moving speed is below a threshold, then such frames can
panning and scanning together with a content based video be removed. It is called time domain seam carving in the
retargeting [9]. Then content resizing based method of the proposed method.
well-known seam carving is proposed [10]. Furthermore,
video carving method is proposed [11] together with The following section describes the proposed method
acceleration algorithm for video carving [12]. These methods followed by some experimental results with video contents.
are referred to “the conventional seam carving method (See Finally, conclusions with some discussions are described.
Appendix).
3
https://www.adobe.com/
4
1 https://www.pyimagesearch.com/2017/01/23/seam-carving-with- https://www.gimp.org/
5
opencv-python-and-scikit-image/ https://www.digikam.org/
6
2 https://scikit-image.org/ https://imagemagick.org/index.php
143 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
144 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
7 8
https://en.wikipedia.org/wiki/Bhattacharyya_distance https://algorithm.joho.info/image-processing/haar-like-feature-value/
145 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
146 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The number of removed frames: 22 frames out of 132 CvHaarFeature, CvHaarClassifier, CvHaarStageClassifier,
frames so that 11/61 of data compression ratio is CvHaarClassifierCascade.
accomplished which is corresponding to the compressed data
volume of 5654KB. On the other hand, the results of moving Then the energy concentration in the other image regions
picture data compression for test scene number 2 is as follows, is calculated for seam carving. Poor energy regions, then, are
removed after that. This procedure is the conventional seam
Image size: 240 by 360 pixels carving method. An example of the conventional image space
seam carving is shown in Fig. 11. In this case, 400 by 300
Data volume a frame: 253KB pixels of the original image are resized in 200 by 300 pixels. It
Threshold: 0.0435 is also possible to determine remaining objects and removing
objects. In this example, the green portion of image would like
The number of removed frames: 47 frames out of 182 to be remained while the red portion of image would like to be
frames so that 47/182 of data compression ratio is removed. Thus, image is resized results in image data
accomplished in this case which is corresponding to the compression.
compressed data volume of 11891KB.
There are some web sites which allow space domain seam
The conventional video seam carving requires much carving. Using web site provided seam carving tool, image
shorter time for play the compressed video with a little bit resizing based on space domain seam carving can be done.
funny impression. Meanwhile, the proposed time domain Fig. 12 shows an example of seam carving image resizing
seam carving requires the completely same time for play the through the web site.
compressed video without any defect because the removed or
carved frames are replaced with the previous frame in
concern.
IV. CONCLUSION
It is confirmed that the proposed video seam carving
method is effective; in particular, time domain seam carving
achieves around 1/5-1/6 of data compression ratio. It is
obvious that the data compression ratio depends on the
changes in the moving pictures. Once the object is detected
with OpenCV library through training with the objective
moving picture itself, then time domain seam carving is
performed effectively.
The conventional image space domain seam carving by
resizing algorithm shows 1/2-1/10 of data compression ratio
depending on redundant image areas in the background in the
image. Therefore, 1/10-1/600 of significantly high data
compression can be achieved with a slight image degradation
Fig. 11. Example of the Conventional Image Space Domain of Seam Carving.
if the proposed space and time domain seam carving data
compression is applied.
The specific feature of the proposed time domain seam
carving is that it requires the completely same time for play
the compressed video without any defect because the removed
or carved frames are replaced with the previous frame in
concern.
Further experimental studies are required for validation of
the proposed seam carving method in time and space domains.
Appendix: Conventional Space Domain Seam Carving
Firstly, seam carving is assumed to be applied to the video
contents in image space domain already. In the space domain
seam carving, objected image portion is extracted first. The
method for object extraction is used to be based on OpenCV
library [13]. How-to build a cascade of boosted classifiers
based on Haar-like features is provided by OpenCV library of. Fig. 12. Example of Seam Carving Image Resizing through the Web Site.
147 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Space domain seam carving allows image resizing with 4) Go back to (2), and refrain the process (3), then the
some intentional conditions. From the web site of Dr. Arial seam carving portion is extracted referencing to energy list
Shamir's paper derived PHP top page, Micro Soft Windows and m list like as shown in Fig. 15.
version of space domain seam carving tool is created.
In the above figure, green line shows the highest energy.
The algorithm is as follows: Low energy pixel which is situated at the far from the highest
1) Get the original image together with the image size energy pixel is removed by seam carving. It looks like an
2) Get the color index of each pixel in the image onion pealing. Other examples are shown in Fig. 16.
3) Calculate energy list using Euclidian distance between Fig. 17 shows an example of image resizing with
the color index of the pixel in concern and the four intentional condition on whether remove the image portions.
neighboring pixels shown in Fig. 13. In this case, the green rectangle in Fig. 17(b) shows the image
4) Take an average over Euclidian distance between pixel portion which must be remained while the red rectangle is the
#1 and #4 and that between #1 and #5 results in a distance in image portion which must be removed. Thus, resultant image
of Fig. 17(c) is reduced with space domain seam carving.
vertical direction.
5) Take an average over Euclidian distance between pixel
#1 and #2 and that between #1 and #3 results in a distance in
horizontal direction.
6) Replace the pixel value of concern (#1) with
summation of the distances in vertical and horizontal
directions
Then energy list is converted to the m list as follows:
①'=①+min(⑦,⑧)
Fig. 13. Calculate Energy List using Euclidian Distance between the Color
②'=②+min(⑦,⑧,⑨) Index of the Pixel in Concern and the Four Neighboring Pixels.
⑥'=⑥+min(⑪,⑫)
⑧'=⑧+min(⑬,⑭,⑮)
Where the pixels in the m list is aligned as shown in
Fig. 14.
Thus, y pixels in the m list are calculated. To identify the
pixels for seam carving, the following algorithm is applied to
the m list,
Fig. 14. Pixels in the m List is Aligned.
1) Extract minimum value of pixel in the calculated
max_y
↓
2) From the max_y-1, the following calculation is made
m_list[x][max_y-1] energy_list[x][max_y-1]
↓
3) The result is corresponding to the following two
candidates:
m_list[x-1] [max_y-2], m_list[x][max_y-2],
m_list[x+1] [max_y-2] Fig. 15. Seam Carving Portion is Extracted Referencing to Energy List and m
List.
148 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
149 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
ACKNOWLEDGMENT [11] Chen B., and Sen P., Video carving, In Short Papers Proceedings of
Eurographics. 2008.
The author would like to thank Mr. Hiroya Inoue for his [12] Chen-Kuo Chiang, Shu-Fan Wang, Yi-Ling Chen, Shang-Hong Lai, Fast
effort through the experiments of seam carving programming. JND-Based Video Carving With GPU Acceleration for Real-Time
Video Retargeting, IEEE Transactions on Circuits and Systems for
REFERENCES
Video Technology,, 19, 11, 1588-1597, 2009.
[1] Kwatra V., Schodl A., Essa I., Turk G., and Bobick A., Graphcut
[13] The Open Computer Vision Library has more than 500 algorithms,
textures: image and video synthesis using graph cuts, ACM Trans.
documentation and sample code for real time computer vision. Tutorial
Graph. 22, 3, 277–286, 2003.
documentation is in O'Reilly Book: Learning OpenCV
[2] Image Retargeting was invented by Vidya Setlur, Saeko Takage, http://www.amazon.com/Learning-OpenCV-Computer-Vision-
Ramesh Raskar, Michael Gleicher and Bruce Gooch in 2005 Library/dp/0596516134
http://en.wikipedia.org/wiki/Seam_carving.
[14] Bhattacharyya, A., On a measure of divergence between two statistical
[3] Boykov Y., and Kolmogorov V., An experimental comparison of min- populations defined by their probability distributions. Bulletin of the
cut/max-flow algorithms for energy minimization in vision, IEEE Calcutta Mathematical Society 35: 99–109, 1943.
Transactions on Pattern Analysis and Machine Intelligence 26, 9, 1124–
[15] Starting with version 3.7, POV-Ray is released under the AGPL3 (or
1137, 2004.
later) license and thus is Free Softwar http://www.povray.org/download/
[4] Wang J., Bhat P., Colburn R. A., Agrawala M., and Cohen M. F.,
Interactive video cutout, ACM Trans. Graph. 24, 3, 585–594 2005. AUTHOR’S PROFILE
[5] Lombaert H., Sun Y., Grady L., and Xu C., A multilevel banded graph Kohei Arai, He received BS, MS and PhD degrees in 1972, 1974 and
cuts method for fast image segmentation, In Proceedings of the Tenth 1982, respectively. He was with The Institute for Industrial Science and
IEEE International Conference on Computer Vision (ICCV’05), vol. 1, Technology of the University of Tokyo from April 1974 to December 1978
259–265, 2005. also was with National Space Development Agency of Japan from January,
[6] Liu F., and Gleicher M., Video retargeting, automating pan and scan, In 1979 to March, 1990. During from 1985 to 1987, he was with Canada Centre
MULTIMEDIA ’06: Proceedings of the 14th annual ACM international for Remote Sensing as a Post Doctoral Fellow of National Science and
conference on Multimedia, ACM, 241–250, 2006. Engineering Research Council of Canada. He moved to Saga University as a
Professor in Department of Information Science on April 1990. He was a
[7] Kohli P., and Torp P. H. S., Dynamic graph cuts for efficient inference councilor for the Aeronautics and Space related to the Technology Committee
in Markov random fields, IEEE Trans. Pattern Analysis and Machine of the Ministry of Science and Technology during from 1998 to 2000. He was
Intelligence (PAMI) 29, 12, 2079–2088, 2007. a councilor of Saga University for 2002 and 2003. He also was an executive
[8] Rav-Acha A., Pritch Y., Lischinski D., and Peleg S., Dynamosaicing: councilor for the Remote Sensing Society of Japan for 2003 to 2005. He is an
Mosaicing of dynamic scenes, IEEE Trans. Pattern Analysis and Adjunct Professor of University of Arizona, USA since 1998. He also is Vice
Machine Intelligence (PAMI) 29, 10, 1789–1801, 2007. Chairman of the Science Commission “A” of ICSU/COSPAR since 2008 then
[9] Wolf L., Guttmann M., and Cohen-or D., Nonhomogeneous content- he is now award committee member of ICSU/COSPAR. He wrote 37 books
driven video-retargeting, In Proceedings of the Eleventh IEEE and published 570 journal papers. He received 30 of awards including
International Conference on Computer Vision (ICCV ’07), 1–6, 2007. ICSU/COSPAR Vikram Sarabhai Medal in 2016, and Science award of
Ministry of Mister of Education of Japan in 2015. He is now Editor-in-Chief
[10] Shai Avidan Ariel Shamir, Seam carving for content-aware image of IJACSA and IJISA. http://teagis.ip.is.saga-u.ac.jp/index.html
resizing, ACM Transactions on Graphics (TOG) - Proceedings of ACM
SIGGRAPH 2007, 26, 3, 10, 2007.
150 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Learning outcomes is one of the important factors student's scores and their ranks. Thus one can anticipate to
to measure student achievement during the learning process. whom student needs extra motivation during the learning
Today’s learning is more focused on problem-solving and process.
reasoning to existing problems than an ordinary problem. Most
exams have been directed to analysis questions for students to E-learning application provides similarity checking on
think and synthesize. As such is troublesome for most students, essay responses. Unfortunately, as such is not quite informative
they are not ready to answer the question, thus, their answers for the teacher such as an unknown source of students copied
almost similar to their friends. This implies that the teacher has work and a similarity relationship between one another. Based
tried to guide students to work professionally and originally. on the observations conducted for about a month on several
However, the Teacher facing difficulty assessing student’s work, campuses in Indonesia, teachers claim that motivation is
especially if the assignments are conducted online without face- compulsory for students, by presenting similarity scores as a
to-face instructions/discussions. To bridge such a gap, the teacher visual report, i.e. graph, chart or diagram.
needs a method or algorithm to measure their rank to encourage
students making an original answer. This research provides a PageRank is a ranking algorithm used by Google to
solution in calculating students' ranks based on the similarity calculate web page rank based on a directed relationship of
score of the essay answers. Pagerank is a ranking algorithm used hyperlinks among others. It uses a power method formula
by Google, this algorithm utilizes a Markov matrix that contains which calculates a Markov matrix to produce Eigen vector. An
the direction of similarity score for each student. These scores are Eigen value is related to its convergences, As a result, Google
computed by the power method until converging. Rank is named it PageRank. PageRank computations take several
displayed to the teacher to review the similarity level of students' hours, depends on the Markov matrix size, to reach convergence
answers. As such is presented a line chart in which the x-axis with initial Eigen vector related to Markov matrix character.
refers to the students and the y-axis depicts the level of similarity.
Ranking computation in Matlab produces an Eigen vector which This research is important to do because lecturers find it
acts as the rank measure. The higher the rank, the more similar difficult to motivate students based on their work as has been
is their answers to others. Hence, students with high ranks to investigated by [1], besides that research objective focuses on
work their answers more seriously ensure their original thoughts. developing a relationship between students based on the
In conclusion, the similarity score matrix using the PageRank similarity score test using the PageRank method. It consists of
algorithm can contribute to the teacher in providing peer parameters: including similarity score and student location
motivation and encouraging student’s internal motivation by (latitude and longitude). Meanwhile, the research scopes are: 1)
presenting the ranked-answers presentation. exams take place is an essay type, and 2) the student’s location
is based on the current local address detected by GPS or
Keywords—Pagerank; learning outcomes; similarity browser with location plugin support.
I. INTRODUCTION II. LITERATURE REVIEW
Online Learning becomes the requirement in industrial era In this section we cover some foundations that grounds this
4.0 to gain the easiest and fastest way of the learning process research consisting of student ranking, plagiarism among
based on technological development. Teachers need it to get students, students’ motivation and page rank method.
the student's exam results promptly and manageable in digital
format. An essay test is still the best method to assess a A. Student Ranking
student's analytical ability to the given problem. The students Vieira (2015) claims the superiority of the strategic value is
to practice his/her skills to analyze and synthesize, as a way to given by the academic ranking process as globally and
develop their knowledge. influences academic quality in scientific activities c. This
In the modern learning goal, a teacher administers an exam article accentuates that student rank has a positive impact to the
at the end of the semester and she or he reviews those exams quality and learning process. Unfortunately, the author does
accordingly, often the teacher finds it difficult to identify the not mentions the cause and effect in a case where a student was
originality of student response and especially their similarity ranked at the low level. As such should lead to the strategic
score. Teachers need an alternative method to measure action to be taken, in which the paper is also lacking of.
151 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Furthermore, Keller (2016) explains that in order to teacher motivated them regarding this plagiarism, but the
improvinge students' intelligence affective factor takes control, results of this study were quite clear.
the teacher must have both knowledge and the right strategy to
motivate students [2]. Affective intelligence and intellectual Khlifi (2017) conducted a simulations that the increasing
intelligence are inseparable to ensure students achieving good quality of the learning process can be achieved by doing
learning outcomes. authentication security during the task process, so the
possibility of plagiarism is minimal [10].
Surveys carried out by Buckley (2016) reveals that
gamified learning has a good impact for the student’s learning Sprajc (2017) in this study states that plagiarism is
process. Rosyid (2018) also reveals that game-based learning currently using technology that facilitates copy-paste and is
strategy can creates a fun-learning environment. However, it is transferred with students with low motivation to study well.
inevitable that participation may varies depending on intrinsic Such a lacking motivation is not only due to plagiarism, but
or extrinsic motivation of the student concerned. The results of can be the product of a poor teaching method. Another research
these study are interesting to analyse by educators, the fact that finds that students who spend time on the internet has no
students at any level of education can take part in such learning relationships to plagiarism scores [11]. Technology is a tool for
methods[3]. But it is not discussed if students dislike a learning facilitating goals, but plagiarism is more about self-awareness.
material, to which strategy the teacher must apply to present Therefore technology should be the driving force for students
the motivation. to learn, where they can find sources of knowledge such as
from the internet.
Marks (2017) said that education process which includes
the effectiveness of learning will affects the quality of learning Kashian (2015) discusses that plagiarism is an ethical act
and increases learning achievement by factors such as course carried out by students. This happens even though the teacher
recommendations, student skills, and behavior detection [4]. has provided advice and guidance to not perform such an
action. But students, mostly, kept doing it. The results of this
Denning (2018) said students with the top three academic study indicate that students with awareness about plagiarism
rank, potentially obtains high test score in the coming exams. have a low plagiarism score. A teacher can help students to
Even, they are found to be easily enrolled the college level. avoid plagiarism by using Turnitin application to check the
This is influenced by the quality and ranking of the school [5]. results of his work. If there is found plagiarism, students are
Students in the higher-ranked schools have higher average expected to act accordingly, for instance rephrasing the
exam score than students from lower-ranked schools. statements in his own words [12]. In some countries this
method is not necessarily applicable, due to application
However, Murphy (2018) explained that the potential licenses and fees. Some developing countries built their own
benefits of schools that have high rankings will be influenced plagiarism detection system and fully supported by the
by the confidence ranking of students who are poorly ranked. government.
With the influence of this ranking, investment in education will
decline [6]. Meanwhile, student self-motivation is still a new Ba (2016) conducts research on 681 student articles at one
research area. Unfortunately, the author did not explain that of the universities in Vietnam using Turnitin and found that the
good school rank is influenced by students rank well. And level of plagiarism in that campus was higher than outside
there is no relationship between low-ranked students and Vietnam with a percentage of 29.06%. Plagiarism will lead to
school’s rank. In facts, there are other factors, involved as well low academic value. In contrast, it will produce a positive
such as school’s accreditation. effect when done honestly, although this task can take some
times. So, it can be concluded that involving a plagiarism
B. Plagiarisme Over Students detection tool is very helpful in developing campus policies
According to Hu (2015), rules that are made based on related to this [13]. In addition, the study did not explain that
detection and focus on giving penalties are more effectively the acquired plagiarism included self-plagiarism or not. This is
applied to Chinese students. On the contrary, based on other because when students writing articles, they sometimes copy-
findings, it shows that an education-based approach that paste statements from another section of the same article.
motivates them is better than giving a punishment [7].
C. Students Motivation
Cronan (2018) mentions behavior patterns, morals, Gianna (2017) provides an interesting picture that the
attitudes are the biggest factors that influence a person to teachers’ motivation is actually contagious to their students. On
academic integrity especially when not to do plagiarism and the other hand, a teachers who is lacking teaching motivation
share homework. The author gives an overview of 33% and can still improve students achievement [14]. This finding
35% of students working on academic task do plagiarism or should enforce schools to encourage teachers teaching
sharing homework [8]. The author did not clearly state what motivations. As such is beneficial to the school and the people
type of motivation had been given to students and how many involved.
students did plagiarism at the school.
Samir (2014) said that teachers need to understand the
Ehrich (2016) conducted research on 131 Australian students’ motivation when teaching online and it is very
students and 173 Chinese students to compare policies against difficult to get because the interaction is not face-to-face.
plagiarism, there was no difference in plagiarism but rather Teachers need to use another strategies including demographic
understanding and plagiarism behavior [9]. There is no variables and the use of technology to prepare the 21st century
discussion in the paper about the response of students when the generation [15].
152 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Darwis (2016) study conducted on EFL fall students during in determining coaches improvement analysis method for
the 2015/2016 semester shows that, plagiarism conducted by better result [23]. In addition such a method should fit for
students was caused by the overly-difficult task material and finding the best trainer for Olympics athletes or the likes.
students desire to graduate soon. Another cause is the openness
of information stream where students can easily retrieve Nykl (2015) aims to rank authors based on journal impact
contents from the internet [16]. One could prohibit the use of value using PageRank algorithm sourced from citation
internet-connected devices in class. However, it is all driven by tracking. The result of this calculation is used to reward the
the student’s motivation in learning and his self-awareness. author for the former year successful inclusion in the database.
This algorithm solely use pagerank and the number of
Harits (2018) proposed approach in developing a SEG publications by eliminating self-citation [24].
game, named Chem Dungeon, as a case study in order to
demonstrate the effectiveness of media. This research making a Pagerank, which was developed by Google, has been used
good contribution to motivate the student via game learning [17]. in various fields such as biology, computer science, physics,
ecology, chemistry, sports and medicine to find ranks
Rocher (2018) reflects cognitive factor as the important according to the research objectives. In addition, understanding
aspect for a teacher to control his students, which including pagerank charts makes it easy to visually understand the target
student involvement in learning, assignments and others. set of interest [25]. PageRank algorithm selection is based on
Plagiarism is a concern that students are not aware of. The previous research which gives the fact that this algorithm is
paper emphasize on in class learning strategy where the suitable for searching ranking based on direction or
teaching learning process becomes active in motivating contribution from other nodes.
students to avoid plagiarism [18]. However, the author fails to
affirms that the easiest step is to announce plagiarism via a Let is adjacent matrix for graph with vertex
banner. * + and the graph is directed graph which is
probability value of weight ( ) . Suppose that is Eigen
Johnson (2018) finds that plagiarism is a complex problem vector, is Eigen value related to , then power method denote
in a human state neither he is conscious or not, It is compulsory as :
to seriously prevent plagiarism in order to maintain academic
integrity on campus rather than penalizing plagiarists [19]. (1)
Johnson however, does not assert the prevention properly and Iterate that formula until convergence.
the method to motivate students to get their hands off
plagiarism. III. METHODOLOGY
D. PageRank In this research, we apply several procedures elaborated in
Amjad (2018) models an academic object ranking, the following section.
including articles, journals, and conference proceedings. This In general, out method refers to causal research where the
ranking is based on the method of link analysis based on problem has been clearly defined in the beginning. The main
publication, number of citations, author's position, influence of problem is how to calculate student ranking based on similarity
the co-author, and topics in the scientific work. Each parameter score using power methods accurately.
influences rank scores to some degrees [20]. Link analysis is
the basis of the page rank algorithm that uses the markov Based on Fig. 1, there are several steps to follow
matrix. The weight of each citation is calculated based on the (1) preparing dataset, (2) conducting primary data, (3) data
count and the list of references. analysis, (4) running power method and (5) ranks report. The
following subsections elaborate each step in details.
Massucci (2019) use pagerank algorithms to study
Record student position on
scientific articles in a campus. Researchers believe that first login
Survey on campus
increasing a rank scores can be done by increasing academic Preparing the
Open e-learning application
alternative is conducting campus-to-campus citations and Generate ranking based Ranking Report
Methodology
on eigen vector
collaborates in the research community.
Find max, min, quartile
Gao (2016) explain that Pagerank index is the reflection of Writing power method code
on matlab
the popularity of publications, in the form of scientific works Importing dataset Find variance, if it has higher
value then repeat conducting
based on the article collection database. The author states that Running Code 1 Data Analysis
primary data
153 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
154 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
155 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 8. Plagiarism Detection Result on Right Table. Fig. 11. Power Method Code on Matlab.
156 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
157 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[15] M. Samir Abou El-Seoud, I. A. T. F. Taj-Eddin, N. Seddiek, M. M. El- [20] T. Amjad, A. Daud, and N. R. Aljohani, “Ranking authors in academic
Khouly, and A. Nosseir, “E-learning and students’ motivation: A research social networks: a survey,” Libr. Hi Tech, vol. 36, no. 1, pp. 97–128,
study on the effect of e-learning on higher education,” Int. J. Emerg. Mar. 2018.
Technol. Learn., 2014. [21] F. Massucci, D. D.-J. of Informetrics, and U. 2019, “Measuring the
[16] S. Al Darwish, A. S.-I. E. Studies, and undefined 2016, “Reasons for academic reputation through citation networks via PageRank,” Elsevier.
College Students to Plagiarize in EFL Writing: Students’ Motivation to [22] C. Gao, Z. Wang, X. Li, Z. Zhang, W. Z.-P. one, and undefined 2016,
Pass.,” ERIC. “PR-index: using the h-index and PageRank for determining true impact,”
[17] H. A. Rosyid, M. Palmerlee, and K. Chen, “Deploying learning materials journals.plos.org.
to game content for serious education game development: A case study,” [23] Z.-H. Hu, J.-X. Zhou, M.-J. Zhang, and Y. Zhao, “Methods for ranking
Entertain. Comput., vol. 26, pp. 1–9, May 2018. college sports coaches based on data envelopment analysis and
[18] A. R. du Rocher, “Active learning strategies and academic self-efficacy PageRank,” Expert Syst., vol. 32, no. 6, pp. 652–673, Dec. 2015.
relate to both attentional control and attitudes towards plagiarism,” Act. [24] M. Nykl, M. Campr, K. J.-J. of Informetrics, and U. 2015, “Author
Learn. High. Educ., p. 146978741876551, Mar. 2018. ranking based on personalized PageRank,” Elsevier.
[19] E. Johnson, “Situational Cheating Assessment of Motivation (SCAM): A [25] D. G.-S. Review and undefined 2015, “PageRank beyond the Web,”
Model for Understanding Student Plagiarism,” 2018. SIAM.
158 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This paper presents a global optimization approach materials and the design parameters [9]. However, the models
aiming to improve the energy efficiency of electrical machines. are quite complicated, the number of design parameters may be
The process is made on a hybrid stepper motor allowing to high and the search for optimality is carried out through
simultaneously optimize design and command. This approach is optimization algorithms. Nevertheless, it would be absurd to
axed around Pontryagin's maximum principle, which is applied seek for the optimal design of an electric machine without
to a magnetodynamic model based on permeances network studying his command. Therefore, the command is one more
model. The originality of the proposed approach is to obtain in parameter that is added to the optimization problems [10]. To
the same process, the minimization of the energy by optimal overcome this difficulty the machine is optimized for operating
control and the minimization of the energy by optimal sizing.
points and imposing a form of control [11]. For example, by
Keywords—Optimal control; optimal sizing; Pontryagin’s
imposing a sinusoidal current control for a given machine, the
maximum principle; permeances network; hybrid stepper motor; problem is to find the optimum amplitude and angle of the
energetic efficiency control to have a certain torque and speed for a given operating
point. The solution of this type of problem requires significant
I. INTRODUCTION computation time, because of the extended model and the
Currently, improving the energy efficiency of electric optimization algorithms used. Therefore, the difficulty is major
machines is a subject of high interest. Indeed electrical if it is to find the optimal value of the command at any time,
machines are widely used in industries, transportation and and in this case, the calculation time will be dissuasive.
home applications. Thus, electric machines consume the largest In this paper, a method is proposed to solve this difficult
amount of energy in the world (i.e. 46% of global consumption problem. The idea is to merge the two domains and to develop
resulting in about 6040 megatonnes of CO2) [1,2]. a global optimization approach for design and control by
The energy efficiency problem is studied in two different applying optimal control theory [12-16] and nonlinear
areas that require different skills. Firstly, automaticians deal optimization algorithms on magnetodynamic models of an
with this problem as an optimal control issue. They seek to find electric machine. This work is applied on a hybrid stepper
the optimal control that allows minimizing either the energy motor [17-22], to prove the feasibility of the approach on a
consumption, subject to some constraints on the control, and/or realistic case.
the performances. Among the performance constraints, we can The paper is organized as follow; in the first part, a
mention the constraints on the torque or for displacement magnetic model of the machine is developed based on a
problems, the constraints of positioning and speed [3]. Thus, permeances network. This model is coupled thereafter to a
the optimal control theory, which is a part of applied dynamic model that describes the electrical and mechanical
mathematics and automatic, is used in dynamic operation of behavior of the motor. In the second part, an optimal control
the machine to find the trajectory of the command. However, theory is implemented, based on Pontryagin's maximum
the machine models used in the automatic field are coarse principle to this coupled magnetodynamic model. At first, an
models [4-7] that do not take into consideration the geometric optimal control problem is posed. Then the optimality
design parameters or not fully the magnetic phenomena. conditions are exploited to conclude on a Hamiltonian model
Secondly, design specialists tackle the energy efficiency that presents a two point boundary value problem. After that, a
problem as an optimal design issue. Therefore, they seek to numerical method is proposed to solve this problem. Finally,
find the optimal design to achieve the required performance optimal controls are calculated for a positioning problem.
while minimizing also energy consumption. Thus, the models Results are then compared with a classical control. In the third
used are more complex and take into account with more part, a global optimization approach is proposed. The link
accuracy the magnetic phenomena [8], the nature of the between the different models and the resolution loops are
presented. Then a global optimization problem is proposed
159 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
followed by a study on the design parameter influence on the aims to represent flux lines by tubes formed with straight lines
energy consumption of the machine. Finally, the global and arcs and to calculate the permeance of each flux tube as a
optimization problem is solved and the results are discussed. function of displacement. Calculations are performed for a
tooth step, the displacement is assumed linear since the length
II. MODÉLIZATION of the angular displacement of a step is negligible in front of
The study is applied to a two-phase hybrid stepper motor, the rotor radius. Fig. 4 describes the tooth structure and the
with 1024 steps per revolution illustrated in Fig. 1. This motor approximation of the flux tubes for a given position. In this
is composed of two rings; each one has 50-tooth. The two rings figure, g represents the gap length, tw the width of a tooth, lr the
are angularly offset by a tooth half step. They are ring length, Pi the permeances of flux tubes and x is the linear
interconnected by a permanent magnet. The stator has 8 plots displacement.
each having Zs teeth. The motor flux distribution is three- A linear displacement x=2tw is equivalent to a rotation of a
dimensional. mechanical angle θ=7.2°. In [22] the permeance expressions
A. Magnetic Model are available and the results of this method have been validated
by finite element method.
The magnetic model is based on a permeances network
method, which consists in decomposing the magnetic device
into a set of flux sources and passive elements. Fig. 2 shows a
front view and a rear view of the MPPH. Pα, Pβ, Pα’, and Pβ’
represents the permeances between the different stator plots
and the two rotor rings. As the structure is symmetrical, this
model focuses on the half of the machine.
Fig. 3 shows the equivalent magnetic circuit. The
reluctance of iron is assumed infinite and the magnet is
modeled by an ideal flux source Fm. The phases are modeled
by the flux sources Fα and Fβ. Fig. 1. Structure of the Hybrid Stepper Motor.
The magnetic circuit resolution aims to determine the flux
flowing in the branches. Equations are performed using
Kirchhoff's laws; the fluxes generated are multiplied by 2 to
represent the entire machine and by the number of coil, Ns, to
describe the flux seen by the coils. The analytical expressions
of the phases flux fed by currents Iα and Iβ generated are given
by:
Fig. 2. HSM, Front and Rear View.
2 P P' N s2 I P P' N s Fm
(1)
2 P P' Ns2 I P P' N s Fm
(2)
The inductance Lα (resp. Lβ) of the phases α (resp. β) is then
deduced:
L 2 P P' N s2
(3)
L 2 P P' Ns2
(4)
Fig. 3. Equivalent Magnetic Circuit.
As well as the mutual flux between the phases and the
magnet which are expressed by:
P P' Ns Fm
(5)
P P' Ns Fm
(6)
The magnetomotive force expression of a magnet as a
function of the length of the magnet lm and the coercive field
Hc is given by:
Fm lm H c Fig. 4. Tooth Structure and the Approximation of the Flux Tubes.
(7)
An analytic method is used to calculate air gap
permenances in order to have a fairly fast model. The method
160 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Cem (t )
1 2 dL 1 2 dL
I I I
d
I
d t0 0 , t f f
2 d 2 d d d (12) I t0 I 0 , I t f
Equations (8), (9), (10) and (12) allow us to write the I t0 I 0 , I t f
(14)
following state model (13):
In equation (14), represents the objective function to be
dI 1 L 1
dt R minimized. This problem can be solved using the Pontryagin’s
L L maximum principle.
dI 1 L 1
dt R
L L A. Hamiltonien and Costate Vector
d
dt 1 1 2 L 1 2 L The Hamiltonian of the system is given by:
J 2 I 2 I I I Cr
d r 1 L
dt
H V V 1
L RI I V
1 0
1 L
L 1 2 RI I V
V L
0 L
V
0 0 1 1 2 L 1 2 L
0 3 I I I I k
0 J 2 2
(13)
4
with Ω is the rotation speed. (15)
161 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
With ѱ1, ѱ2, ѱ3 and ѱ4 are costate variables, J is the sum of d 1 1 L 1 L
the rotor and wheel inertia, and k is the sum of the viscous dt 1 L R 3 J I
friction coefficient of the rotor and the wheel. d
1 L 1 L
2 2 R 3 I
According to the maximum principle, the costate vector dt L J
must verify the following relation:
d 3 1 I L 1 I L k
dt
L L
H
1 2 3 4
d 1
J
dt I d 1 1 L
2
2 L 1 L 2
4 1 I
L
I
L
2
dt
d 2 H
2
L
dt I 1 1 L 1 L
1 RI
V*
d 3 H L L L
dt
1 1 L
2
2 L 1 L
2
I I
d 4
H L L L 2
2 2
dt 1 1 L 1 L
(16) 2 RI
V*
L L L
These relations give the first optimality condition.
1 1 2 L 1 2 L
2
2 2
B. Optimal Control Expresssion 3 I2 I I I
J 2 2
2 2
2
2
The second optimality condition of the maximum principle
indicates that the optimal control minimizes the Hamiltonian. dI 1 I R L 1 1 V *
dt L L L
Therefore, to find the command expression the sign of the
functions corresponding to each command and derived from H dI 1 L 1 1 *
dt L I R L L V
has been studied. For example, the function derived from H
with respect to Va is given by: d 1 1 L 1 2 L
I 2
I I I k
V 1 dt J 2 2
HV 1
d
V L
(17) dt (20)
For ψ1(1/Lα)>0, HVα is strictly positive, then H(Vα) is
The Hamiltonian model obtained is a two point boundary
strictly increasing and the minimum of the Hamiltonian is
value problem. Indeed, it is necessary to find the initial
reached in Vα=Vmin. Thus, studying the sign of the derivative
conditions of the costate variables allowing bringing the
allowed to obtain optimal command expressions. For Vαmin=-
system from its initial state to the desired final state. As for the
Vαmax we get the expressions of the optimal controls Vα* and Vβ*
final state of the current, it was left free. This implies
are:
transversality conditions on the costate vector. In fact, if the
1 1 final state is free, the corresponding costate vector must be
V min sign 1 if 1 1 equal to zero:
L L
V* 1 t f 0
1
0 if 1 1
L 2 t f 0
(18) (21)
1 1 To solve this kind of problem of boundary condition type,
V min sign 2 if 2 1 the so-called shooting method has been used, which aims to
L L create a function S that takes as inputs the initial conditions of
V
*
the variables and returns the difference between the final state
1
0 if 2 1 obtained and the desired final state and the transversality
L conditions.
(19)
The command with a minimum of energy for this problem 1 t0 1 t f
2 t0 2 t f
involving constraints on the commands is therefore Bang-off-
bang type. In (18) and (19) the command is expressed in terms S :
of the costate variables ѱ1, ѱ2 which in turn are expressed as a 3 t0 t f f
t
4 0 t f f
function of all the parameters of the machine according to the
relation (16). The control is expressed explicitly according to (22)
all the parameters taken into account by the model.
The root of the S function should be determined. This can
The equations (16), (18), (19) and the machine state allow be done with an algorithm based on the newton method, the
(13) us to obtain the Hamiltonian model (20): fsolve routine of Matlab®. However, one must be able to
162 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
min Obj (t ) V
1.2 s
V dt
V ,V 0
With;
5V V 5V
5V V 5V
t0 0 rd , t f 0.5835rd
t0 0rd / s , t f 0rd / s
I t0 0 A , I t f
I t0 0 A , I t f
(23) Fig. 8. Optimal Control Vα*(t).
163 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
taking into account of the inertia. For the classic control, the 0 (24)
position is reached more quickly but the motor in this case
Minimizing the objective function Obj’ is the minimization
consume more energy. The obtained optimal control offers a
of the Obj function. The only difference is that the Obj’
gain of absorbed power of 5,6%.
function would have a smoother evolution and would be more
Fig. 11 presents the evolution of the costate variables ψ3(t) efficient we use a gradient descent.
and ψ4(t) related respectively to the speed Ω(t) and to the
Fig. 14 describes the evolution of the function Obj’ as a
position θ (t). The costate variables ψ1(t) and ψ2(t) will have the
function of lm and lr.
same pace as the switch functions with a difference in
amplitude due to the terms 1/Lα and 1/Lβ in (8) and (9).
164 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
process, the minimization of the energy by optimal control and
min min
lm , lr V ( t ),
V ( t )
0
0.1 s
V V dt
the minimization of the energy by optimal sizing. First, a
magnetodynamic model based on a permeances network was
with : developed. Then the Pontryagin Maximum Principle was
applied to the magnetodynamic model in order to find the
8mm lr 22mm
optimal control minimizing the energy. The application of the
0.5mm lm 1.6mm PMP allows us to explicitly express the command according to
5V V 5V all model parameters and to have a Hamiltonian model that
automatically generates optimal control. The study has shown
5V V 5V
that boundary value problem encountered at the resolution
t0 0 rd , t f 0.157rd level of the Hamiltonian model can be solved by a simple
t f 0rd / s
dichotomy when it is a control problem of an electric machine.
t0 0rd / s , The results showed a gain of 5.7% compared to a conventional
I t0 0 A , I t f control for a given positioning problem. Finally, adding an
I t f
optimization loop on design inputs to give an overall
I t0 0 A ,
(25) optimization approach.
165 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
166 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The Model of Game-based Learning in Fire Safety save lives. Addressing these problems, a widespread
developed for preschool children to educate them in learning fire implementation of teaching and learning using technology was
safety issues. Due to the lack of awareness towards fire hazard, focused to reduce injuries and fatalities among the people
there are few factors that have arisen regarding this issue such as especially children.
children’s ages, experiences and knowledge. The main objective
of this study is to identify the user requirements of preschool In addition, there is a growing need for the preschool
children in developing the Model of Game-Based Learning in children to learn using technology instead of conveying the
Fire Safety. This study involved six preschool children of Tabika information through conventional way of teaching and
Kemas Kampung Berawan, Limbang Sarawak by using User- learning [3]. Subsequently, technologies have been exposed
Centered Design method. The ability of cognitive, behavior and all over the world and spread like wildfire by providing a lot
psychomotor skills are the main aspects to develop the model. of functionalities that contribute in helping people. Therefore,
Thus, to lower the risk of injuries during practical training in the development of interactive technology such as tablet led to
real situation, there is a need to educate them using the continuous innovation of teaching and learning. Apparently,
technology of tablet. Therefore, a prototype has been developed the technology itself stimulates creativity and often draws the
known as APi Game-Based Learning as a platform for children attention of preschool children [4]. However, children are
to learn about fire safety issues. Hence, this APi prototype facing difficulties when using technology because of some
developed to validate the Model of Game-Based Learning in Fire
constraints due to their abilities and growth development.
Safety development for preschool children. Thus, the finding of
Many applications developed to ease them in using technology
the study showed the engagement of children in learning fire
safety through game improved their knowledge, behavior and
and its use in education trained their cognitive and
psychomotor skills. Overall, this study makes an important psychomotor skills [5], [6].
contribution in determining the usability on the level of The implementation of applications that support learning
effectiveness towards preschool children through active learning. helped the children to interact well with the systems. For
example, the application of fire safety education in the form of
Keywords—Game-based learning; fire safety; user-centered
gaming environment. Thus, the interactive learning such as
design; effectiveness
game-based learning enhanced the children to improve
I. INTRODUCTION interaction and learning in fire safety issues [7]. On the other
hand, to provide them clearly on the risk of fire hazard that
Lack of awareness in potential fire hazards as the concern caused a lot of damages and death required an efficient way to
issues bring threat for the children. Addressing this need, the identify their needs.
exposure of fire safety issues on the children led them to learn
survival skills. In fact, fire hazard affects their lives, education II. LITERATURE REVIEW
and abilities. Many cases in Malaysia as reported by Fire
Game-based learning promoted interactive learning, which,
Rescue Department of Malaysia showed the annual report of
derived from the use of computer games in delivering the
fire cases related with houses and buildings [1]. Due to the
educational value [8]. The main purpose of developing
faulty of electricity as the highest factor of deadly fire
educational games and training simulators were using the
occurred every year. Thus, by providing the information of
same technical elements with different purpose of delivering
fire hazard especially to children plays an important role to
information to the users. There were many applications
educate them and save lives as well. The importance of
developed in the form of gaming to educate the children in
learning and teaching fire safety issues should be taken
teaching and learning [9], [10], [11]. Realizing the best ways
seriously to prevent injuries to the children.
to attract the children to stay focus during learning session by
These are the most common problems occurred that related providing multimedia elements such as animation and audio
with fire safety. There were caused by limited fire safety [12], [13], [14].
awareness, less training, fire protection systems were not
Getting feedback from the system while interacting, the
provided and delayed on notifying the fire brigades [2]. Fires
children gave either positive or negative responses and kept on
are always life-threatening and dangerous that risk people to
167 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
168 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. THE PROBLEMS OF EXISTING FIRE SAFETY GAMES TABLE. III. USER REQUIREMENTS
Types of Games Initial Investigation Issues of Interaction (Users) User Requirements 4 years 5 years 6 years
User Interaction
1. User only interacted
1. Limited user
with the limited button. Interface Design
interaction.
2. The interface was quiet
2. Lack of Psychomotor (Fine Motor)
Help Mikey confusing for the users.
information.
Make It Out 3. Less instruction for the
3. Use of language. Cognitive (Knowledge)
users on how to play.
4. Less guidelines
4. Difficulty of Behaviour
provided.
understanding English.
Gaming Elements (Reward,
1. Lack of 1. No guidelines on how to Storyline, Player, Time)
information play. Multimedia Component (Animation
Fire Safety 2. Element of 2. No instructions through
and Audio
Challenge sound or voice voice provided.
used. 3. Difficulty of Genre (Strategy)
3. Use of language. understanding English. Malay Language
The interaction styles issues were identified after the Besides, observation technique defined on observing their
games tested on the preschool children. Following the capabilities of solving the games. Meanwhile, think aloud was
experiment conducted, the data was collected through a process of determining the children’s opinions on
observation and think aloud. As shown in Table II, the engagement and enjoyment of playing the games. Thus, the
children played the games with different skills level. It showed data collected can be implemented to develop the Model of
that different ages of children placed some constraints on the Game-Based Learning in Fire Safety as shown in Table II and
ability of cognitive, behaviour and psychomotor skills when Table III.
being tested by the existing fire safety games. Based on Table III showed that the user requirements
obtained after carrying out the experiment on the existing fire
TABLE. II. ANALYSIS OF PRELIMINARY STUDY
safety games. All the user requirements were focused on
Age of children’s abilities, gaming elements and gaming factors [3],
Help Mikey Make It Out Fire Safety Challenge
Children [6], [15], [19], [22], [25], [26].
1. Understand on how to Analyzing all the user requirements for developing the
play the game but 1. Understand on how to model such as user interaction, interface design, psychomotor
needed more play the games which skills, cognitive, behavior, gaming elements, multimedia
explanation because of showing positive
the language used. response for getting components. However, during the experiment conducted on
2. Response towards the correct answers. the preschool children, they were facing trouble to understand
game was slow because 2. Slow response because English because they were not using English as the main
4 years of the instructions given the interface was quiet language at home and school. All of the preschool children
on how to play were not confusing. participated well in the experiment which, showing the
clear. 3. There were no proper
3. Able to point the answer guidelines to play where
positive responses towards fire safety games. Some of them
using the buttons only music provided. were facing difficulties on interacting with the buttons must be
provided. 4. Difficult to point designed specifically for children to play with ease.
4. Good response on precisely at the objects.
getting rewards. B. Model of Game-based Learning In Fire Safety (Design
Phase)
1. Interested in The Model of Game-based Learning in Fire Safety was
visualisation and audio
in the games.
1. Not really interested in developed based on the combination of Game-Based Learning
playing because no Model [22], Fire Safety Model [27] and the user requirements
2. Needed more
voice instruction
5 years explanations because of
provided.
obtained from the preliminary study as shown in Fig. 3.
the language used was
2. Able to point directly at Game-based Learning Design Model:
difficult to understand.
the objects.
3. Good response on
getting rewards. 1) User: Preschool children at the age of four to six years
old are required to testify the effectiveness of fire safety game.
1. Able to complete the
1. Enjoying the game 2) Device: Tablet is a tool used as the device to test the
when getting reward. game which, showed the compatibility with the children’s fine
games after instruction
2. Able to control finger
given. motor skills and cognitive.
6 years movements precisely.
2. Slow response because
of the language used in
3. Slow response because 3) APi Fire Safety Game: It consists of game elements
of no instructions such as rewards, player, storyline, feedback, time, multimedia
the game.
provided.
components to improve motivation, enjoyment and
interactivity. Besides, interface design is focusing on menu
Strategic Research Grant of KRA-2018-025, Faculty of Information
Science & Technology, Universiti Kebangsaan Malaysia.
169 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
driven to guide the children in using fire safety game by Software Contents: There are three aspects will be
providing animation and sound. The voiceover was using evaluated after testing the games that include behaviour,
Malay language. cognitive and psychomotor. Meanwhile, the effectiveness is
4) Prototype: It consists of low-fidelity prototype and the factor of usability used to validate the Model of Game-
Based Learning in Fire Safety.
high-fidelity prototype.
The model consisted of three missions that required the
User Device users to complete. Mission 1 needed the users to identify the
USER causes of fire accidents such as inflammable substances while
Preschool INTERACTION Tablet
Mission 2 conveyed the information on how to use fire
Children Android
Platform extinguisher and the way to escape from fire in a trap house.
Meanwhile, Mission 3 required the users to identify whom
they should rescue if fire accident happened. All these
APi Fire Safety Game missions provided for the users to learn and know the
importance of fire safety issues. Every missions tested on the
users would affect their cognitive, behaviour and psychomotor
Game Elements Interface Design skills. Thus, all the missions created based on the learning
Rewards module of Tabika KEMAS.
Player Menu-driven
Storyline
All the game elements were obtained based on suitable
Feedback Time elements needed for preschool children [8]. During
Multimedia Components preliminary study showed that audio and animation were the
Animation
Game Genre: Strategy most important elements for them to engage while playing the
existing fire safety games. These game elements improved the
Audio children’s motivation, enjoyment and playing interactive game.
Motivation
(Voiceover
in Malay This APi fire safety game was developed in 2D which
Enjoyment language) categorized in strategy and offline game [26]. Besides, the
interface design was based on menu-driven to ensure the users
Interactive understand the game flow of playing.
In addition, every mission in this game was following the
module of learning at Tabika Kemas Kampung Berawan,
1
Limbang Sarawak. The missions developed were suitable for
2
Low-fidelity Prototype High-fidelity Prototype the preschool children in terms of the gameplay and easy to
understand the contents. This was because the APi fire safety
Evaluation of APi Usability evaluation of
Interface Design
game was using Malay language as the medium of interaction
APi
Cognitive Observation
between the users and game. All the instructions were given
Walkthrough Think aloud
by using audio instead of text because the limitation of the
preschool children in reading.
There were two types of prototypes developed based on
EFFECTIVENESS the model. It consisted of low-fidelity prototype and high-
fidelity prototype. In order to evaluate the low-fidelity of APi
prototype, the cognitive walkthrough method was conducted
on experts. This method used to improve the interface design
Software Contents of the prototype. While, to evaluate high-fidelity prototype
were using observation and think aloud on the real users, the
Cognitive Behaviour
preschool children. This final evaluation used to evaluate the
Psychomotor
effectiveness of APi fire safety game tested on them.
Testing the Determine the
The effectiveness of usability was evaluated based on
users’ abilities of the
knowledge. users’ fine Responses of cognitive, psychomotor and behavior of preschool children.
motor skills. the users Through the APi fire safety game, the children’s abilities were
Instruction of towards fire tested in solving the missions with the use of tablet technology.
gameplay in Touch safety before, The users’ fine motor skills would be tested too to ensure that
Malay language. interaction during and
they could use touch interaction easily. With the use of APi
style. after playing
Activities the game. fire safety game, their behaviors were evaluated before, during
(Mission 1, Direct touch and after playing the game.
Mission 2, used during
Mission 3) gameplay. C. Developing APi Prototypes
In this section discussed on the prototype development
Fig. 3. Model of Game-Based Learning in Fire Safety for Preschool which consisted of two processes, low-fidelity APi prototype
Children. and high-fidelity APi prototype. It was used to be named as
170 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
APi because of the fire theme that related with fire safety D. Cognitive Walkthrough Method
issues. These prototypes were developed based on the model Extending to this low-fidelity APi prototype development,
shown on Fig. 2 for preschool children and suitable with their a Cognitive Walkthrough method was carried out to improve
ages, skills and knowledge. Both prototypes were designed the weakness of the interface design [18], [28]. This process
using Adobe Photoshop and Unity 2017. APi prototypes involved experts that consisted of two lecturers, two gamers
emphasized more on the interaction style issues that needed to and two graphic designers. Those experts needed to perform
solve the research problems. Table IV and V showed the the tasks given to them in 20 minutes without discussing with
processes on developing the APi prototypes. other participants.
A low-fidelity prototype was not a fully functioning The experts were given the tasks to be completed during
system developed for the users. This process was showing the experiment along with the low-fidelity APi prototype. The
early stage of development to test the functionality and tasks were shown below:
interface design that suitable for the real users. Table IV
showed the interface design of low-fidelity APi prototype. 1) The participant needs to press the button “MULA” at
the APi main interface to start the game.
TABLE. IV. LOW-FIDELITY PROTOTYPE 2) There will be missions on the next screen which,
consists of “MISI 1”, “MISI 2” and “MISI 3”. The participant
APi INTERFACE DESIGN DESCRIPTION
needs to test the functionality of HOME button to go back to
the main interface.
The main APi interface which 3) Next, the participant needs to choose “MISI 1” as the
required the players to choose starting mission. The instruction will be given through audio
“MULA” or “KELUAR.
before the mission has started. The participant needs to test the
MULA : Start the game functionality of the button to proceed to the next screen.
KELUAR : Exit the game
APi Main Interface 4) The participant starts the “MISI 1” by pressing the
“MULA” button and plays the game. Then, there are left and
right button provided to move the character after the
instructions given through audio.
The player can choose the three 5) After completing the “MISI 1”, the participant can go
missions which are Mission 1, to the mission screen by hitting the “OUT” button provided at
Mission 2 and Mission 3. the right side.
6) The participant will choose “MISI 2” for the next game.
Missions’ Screen The instructions will be given through audio to guide the
participant. There will be RIGHT and LEFT button provided
The instructions given to the players to go to the next page.
by using audio and animation on 7) The participant will be given instructions on how to
how to play. The players need to
identify the flammable substances.
play by using UP, DOWN, LEFT and RIGHT button provided.
8) After completing the “MISI 2”, the participant can go
The gameplay started by pressing
to the mission screen by hitting the “OUT” button provided at
the left and right button to catch the
Mission 1: Gameplay objects. the right side.
9) Next, the participant needs to choose “MISI 3” and the
instruction will be given through audio before the mission has
The instructions given to the players
by using audio on how to play. The started. The participant needs to test the functionality of the
players need to think on how to button to proceed to the next screen.
escape from the house with fire. 10) The participant starts the “MISI 3” by pressing the
There were four buttons provided “MULA” button and play the game. Then, there are RIGHT
for the players to move the and LEFT button provided to move the character after the
Mission 2: Gameplay character. instructions given through audio.
11) After completing the “MISI 3”, the participant can go
The instructions given to the players to the mission screen by hitting the “OUT” button provided at
by using audio on how to play. The the right side.
players needed to identify whom
they should rescue from fire.
12) The participant will be rewarded by giving the badges
after all the missions completed. Then, the participant needs to
Pressing the right and left button to
move the character in catching the
go back to the APi main screen by hitting the “OUT” button.
Mission 3: Gameplay objects targeted. 13) At the APi main screen, the participant can choose
“KELUAR” to exit the game.
171 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
As shown in Table V was the analysis of Cognitive TABLE. VI. HIGH-FIDELITY PROTOTYPE
Walkthrough conducted on the expert users.
APi INTERFACE DESIGN DESCRIPTION
Through the Cognitive Walkthrough method tested on the
experts, there were some interaction and interface designs
needed to be improved in low-fidelity APi prototype. Thus, it The main interface design added with the
involved with the controller buttons, font and size, animation images of fire to let the users know that
of the buttons, background image, consistency of button’s they are learning fire safety game.
position and visibility of the buttons.
APi Main Interface
The analysis showed that APi prototype had the features
that suitable for the preschool children. The missions provided The background theme related with Fire
for them can be easily played with minimal supervision where, and Rescue Department of Malaysia. The
position of “MISI 1”, “MISI 2”, “MISI
the children did not need to read the instructions of the game 3” changed along with the font type and
due to their capabilities in reading. This was because every size.
instruction given to them through the audio.
MISI 1: Mission 1, MISI 2: Mission 2,
As shown in Table VI were the high fidelity of APi Mission 1: Gameplay MISI 3: Mission 3
prototype after cognitive walkthrough method was conducted
on six experts.
The fire images added along with
“Bahan Mudah Terbakar” to educate the
TABLE. V. ANALYSIS OF COGNITIVE WALKTHROUGH users clearly. Besides, the position of
Next and Previous buttons changed to
CRITERIA DESCRIPTION AGREE DISAGREE avoid confusion among the users.
1. The background Mission 2: Gameplay
colour. 50 50 The controller button changed on the
Background
2. The use of design and position. For the design, the
Theme
images as 50 50 controller buttons were fixed to ensure
background. the consistency in every mission.
1. Font size 50 50 Meanwhile, there were differences
Fonts between the positions of controller
2. Font type 50 50
buttons and Next, Previous buttons to
Mission 3: Gameplay avoid confusion for the users.
1. Size of button
50 50
Buttons 2. Consistency of
33 67
(“MULA, shape and size
KELUAR”) 3. Animation style The background theme changed based on
100 0
of button the mission to escape from fire in house.
The position of Next button was fixed
1. The use of 33 67 too.
button to move
Controller character Mission 2: Instruction
Button 2. Consistency of 17 84
shape and size.
3. Position of button 0 100 The controller buttons added with UP
1. To deliver the and DOWN to control the player
Icon 83 17 movements. The buttons were fixed to
information.
ensure the consistency in design and
1. Referring to the 83 17 position.
player Mission 2: Gameplay
Character
2. The use of colour 50 50
3. Size of characters 100 0
IV. RESULTS
1. The way of
playing and The development of the high-fidelity APi prototype was
Game Goals 100 0
delivering tested on the real users that were preschool children. At this
information. evaluation of the APi prototype to evaluate the effectiveness
1. The position of
of the Model Game-Based Learning in Fire Safety developed.
100 0 The evaluation phase involved six participants of preschool
menu
Menus 2. The use of colour children by using two techniques of observation and think
83 17
3. Animation style
100 0 aloud.
of button
These techniques used to identify the reactions of the
1. Missions 100 0 preschool children either they were able to complete all the
2. Reward 100 0 missions in specific time given. The ways of responding to the
Features
3. Time 100 0
4. Score 100 0
game were the main part to evaluate the effectiveness of APi
fire safety game tested on them.
172 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
As shown in Table VII, the results showed on the score TABLE. VIII. HIGH-FIDELITY PROTOTYPE
obtained while playing the APi fire safety game. Age of
Cognitive Psychomotor Behaviour
Meanwhile, Fig. 4 showed the three main aspects Children
investigated in this study, which were cognitive, psychomotor, MISSION 1:
and behaviour of the preschool children. The usability testing MISSION 1: A bit slower to
was conducted to validate the effectiveness of the fire safety Some of them control the game
were unable to controller buttons.
game-based learning model for preschool children. The MISSION 1:
memorize the But still able to
percentage of cognitive aspect showed only 50% of preschool inflammable control the left and Giving the positive
children at the age of 4 could accomplish Mission 1 due to the right buttons but response towards
substances easily.
ability of memorizing the inflammable substances was still at facing difficulties inflammable
But still able to
with the speed of substances.
the low level. 100% obtained for both 5 and 6 years ol complete the
mission to find moving objects. MISSION 2:
d preschool children where, they could solve the problems
inflammable Shouting loudly to
completely. MISSION 2: seek for help when
substances. Able to control the
For psychomotor aspect showed that all of preschool 4 years facing danger.
MISSION 2: left and right Happy to complete
children were able to control their finger movements while Able to complete buttons easily. the game.
playing the game. They could fix the hand-eye coordination the mission to MISSION 3:
escape from the MISSION 3:
although 4 years old children were having difficulties in A bit slower to Giving the positive
burning house.
handling the speed of the objects. Due to the growth control the game response to save
development of fine motor skills, the children were trying to MISSION 3: controller buttons. lives and the
Able to complete But still able to
play with the buttons provided. the mission to control the left and
engagement while
playing the game.
save lives of right buttons but
Based on the percentage showed that 50% of the preschool
human and facing difficulties
children at the age of 4 could finish the missions. They were animal. with the speed of
showing positive response and engagement during playing moving objects.
session. However, some of them was easily distracted due to
the frustration of the incapable getting the objects correctly. It MISSION 1:
The engagement of
led the children to give up on playing the game. Other than MISSION 1: the player to
that, preschool children at the age of 5 and 6 years old showed MISSION 1:
Able to control the remember every
the engagement and motivation while playing the game. They Able to complete
buttons easily inflammable
the mission to find
kept on responding to seek for help when facing danger. Thus, inflammable
without substance.
they were able to understand the dangers of fire that led them difficulties. MISSION 2:
substances.
to injuries. Their behaviours were observed before, during and MISSION 2: Easy
MISSION 2: Afraid to solve the
after playing the game. Able to control the game because of
to complete the
buttons easily to the shouting voice
5 years mission to escape
Therefore, Table VIII showed the analysis of the usability from the burning
move the but still able to
testing evaluation on preschool children at the age of 4 to 6 characters. point out the
house.
years old. Every mission aligns all different activities to test MISSION 3: escape door.
MISSION 3: Able
their capabilities on cognitive, psychomotor and behaviour of Able to control the MISSION 3:
to complete the
the children. buttons easily with Showed the
mission to save
the fast speed on engagement of the
lives of human
catching the player to save lives
TABLE. VII. SCORES OF THE API FIRE SAFETY GAME and animal.
objects. and happy to
Mission R1 R2 R3 R4 R5 R6 complete the
game.
Mission 1 50 50 100 100 100 100
Mission 2 100 100 100 100 100 100 MISSION 1:
MISSION 1:
Know the
Mission 3 50 50 100 100 100 100 MISSION 1: Facing no
functions of
Easy to complete difficulties in
inflammable
the mission to find handling touch
substances and
Usability Evaluation inflammable interaction with
showed the
substances. the controlling
120 engagement while
MISSION 2: buttons.
playing.
100 Easy to complete MISSION 2:
80 MISSION 2:
the mission to Easy to handle all
6 years Shouting loudly to
60 escape from the the buttons
seek for help to
40 burning house. precisely to move
escape from fire
20 MISSION 3: the character.
and completing the
0 Easy to complete MISSION 3: game easily.
Cognitive Psychomotor Behaviour the mission to Facing no
MISSION 3:
save lives of difficulties in
Showed the
4 years 5 years 6 years human and handling touch
engagement while
animal. interaction with
playing to save
the buttons.
Fig. 4. Evaluation of usability Testing. lives from fire.
173 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The results showed that the children at the age of four Based on BIM (building information modelling) that
years old have difficulties in controlling the speed of moving supported by virtual reality and serious games exposed the
objects and game controller buttons. But they still managed to users on awareness of emergency training [14]. By providing
play the game till the end. For Mission 1 and Mission 3, either real-time fire evacuation guidance let the users to understand
they were able to use touch interaction by using only one evacuation process. Three main aspects were focused which
finger or two fingers on both hands. These missions let the were real-time two ways information updates, real-time
children to focus by using hand-eye coordination activities. evacuation route and real-time location of building. Through
the activities provided for the users to accomplish, their
For Mission 2, most of them completed the mission with behaviours were taken seriously towards the experiment. The
ease by escaping from the burning house. It required the research highlighted the effective way of actions for human
children to think as fast as they could to find the escape way. behaviour in emergent situations. The Model of Game-Based
Some of them were shouting very loudly to seek for help. For Learning in Fire Safety focused on the real-time game where it
Mission 3, it showed that the engagement of playing the game provided users to accomplish the mission within the limited
improved their attention span. time given. Hence, by playing the game would improve the
They tried to solve the mission with limited time. Thus, by children’s ways of thinking and right actions taken.
scoring points in the game kept them motivated to play more. In this Model of Game-Based Learning in Fire Safety
The children were able to hear the instruction given clearly research highlighted the three main aspects of preschool
and followed the gameplay. All these missions provided children which were cognitive, psychomotor and behaviour
helped to stimulate their brains in solving problems if they towards fire hazard. Through gaming environment attracted
were facing the fire hazard situation. Through this APi fire them to engage well with the missions provided in APi
safety game helped the preschool children to improve awareness prototype. Concerning the fact that children have limited
on fire safety issue at the early age. So that they noticed on how
capabilities compared to adults, all the missions created were
life-threatening fire is. It may cause injuries and death. By suitable for their ages, knowledge and skills [20]. Thus, they
conveying the basic knowledge of fire safety on them through
could solve the missions in given time to test their skills in
game environment, it helped them to engage more in learning. handling the technology of tablet. By using the touch
Therefore, it showed the effectiveness of the Model Game- interaction could identify their psychomotor of fine motor
Based Learning in Fire Safety developed. The preschool skills and capabilities. APi prototype provided learning
children understood the gameplay and learned the issue of fire through game environment to verify the model developed.
safety with full of excitement. Based on the APi Game-Based The usability evaluation showed the effectiveness of APi
Learning helped to improve their cognitive, psychomotor and prototype tested on the preschool children to educate them in
behaviour towards fire safety issues. Learning fire safety learning fire safety issues. The preschool children were able to
issues through gaming environment improved their motivation complete the mission with minimal supervision. On the other
too. Thus, game-based learning allowed better understanding hand, some of them were facing difficulties in controlling the
of extremely dangerous effects of fire on children. In fact, buttons. It happened because of the speed of the moving
educational game helped in promoting fire safety awareness. objects. Overall, all the preschool children showed positive
V. DISCUSSION response towards fire hazard and knew the basic skills to
escape from danger.
Time was a crucial moment in handling fire situation. In
fact, causing the risks of injuries and death could be reduced APi prototype was developed based on the user
by providing the fire training simulator. Virtual reality based requirements of the Model of Game-Based Learning in Fire
fire training simulator provided the general public and Safety. Therefore, all multimedia components used such as
firefighters to train them in making decisions and organized audio and animation in developing the APi prototype
responses towards fire safety [5]. By providing activities to eventually helped the preschool children to stay focus while
achieve the goals of effectiveness, the users needed to perform playing. Audio and animation aroused their attentions and
training and experienced fire environments. Apparently, engaged them to play longer. With the instructions given to
evacuation and rescue activities of fire situation at road them by using voice, the preschool children did not need to
tunnels were evaluated on the users. Based on the proposed read the instructions because some of them were still having
framework, the functions and real-time performance of the difficulties in reading progress. While, the use of technology
simulator were verified. Their behaviours on initiating right tablet attracted the preschool children to play because of their
actions were observed specifically in handling real fire situations. ages to explore the surrounding.
Therefore, by accomplishing the missions or activities of Extending to this research, we hoped that the Model of
fire safety helped the users to think as fast as they could in Game-Based Learning in Fire Safety can be used in other
order to rescue themselves. Conveying the information of fire disaster such as flood to guide the preschool children. They
safety without harming the users was important to let them need to be exposed and trained well from the early age as the
aware and initiate effective response towards danger. There prevention from danger situation. This is because the children
were a lot of technologies used to deliver new methods of are highly-risk towards fire hazard that can cause damages and
learning and training the fire safety education such as CAVE death. With this in mind, saving lives are important by doing
and computers [10]. In addition, these technologies used the right actions.
should be compatible with the users’ abilities and skills.
174 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
VI. CONCLUSION [11] Tsai, M. H., Wen, M. C., Chang, Y. L., & Kang, S. C., “Game-based
education for disaster prevention,” AI and Society, vol 30, pp. 463–475,
This study indicates the development of the Model of Nov 2015.
Game-Based Learning in Fire Safety for preschool children. [12] Noorhidawati, a., Ghalebandi, S. G., and Siti Hajar, R, “How Do Young
All the user requirements needed should be discussed more in Children Engage with Mobile Apps? Cognitive, Psychomotor, and
order to train them on fire safety issues. Due to the lack of Affective Perspective,” Journal of Convergence Information Technology,
vol 87, pp. 385-395, July. 2015.
awareness towards fire hazard, there is a need to educate them
in the form of edutainment to lower the risk of injuries. [13] He, Q., Hong, X., Zhao, G., & Huang, X., “An Immersive Fire Training
System Using Kinect,” in Proceedings of the 2014 ACM International
Therefore, the preschool children needed to be exposed Joint Conference on Pervasive and Ubiquitous Computing: Adjunct
Publication, 2014, paper 14, p. 231–234.
well and alerted on how life-threatening fire is. Despite the
design of the Model of Game-Based Learning in Fire Safety [14] Wang, B., Li, H., Rezgui, Y., Bradley, A., and Ong, H. N., “BIM based
virtual environment for fire emergency situation,”, The Scientific World
promoted on the fire safety awareness. Hence, it helped to Journal, pp 22, Aug. 2014.
foster the children’s learning process in their daily lives. [15] Abdul Jabbar, A. I., & Felicia, P., “Gameplay Engagement and Learning
Further study should be directed towards addressing on how in Game-Based Learning: A Systematic Review,” Review of
the model can be improved based on children’s needs such as Educational Research, vol 85, pp 740–779, 2015.
cognitive, psychomotor and behavior aspects. [16] Kamarudin, D., Hussain, Y., Applegate, E. B., & Yasin, M. H. M., “An
Ethnographic Qualitative Study On The Malaysian Preschool And
ACKNOWLEDGMENT Special Needs Children’s Home And School Reading Habits,”
International Journal of Pedagogy and Teacher Education (IJPTE), vol 2,
This work was supported under the Strategic Research pp. 224–234, April 2018.
Grant of KRA-2018-025, Faculty of Information Science & [17] Towers, B., “Children ’ s knowledge of bushfire emergency response,”
Technology, Universiti Kebangsaan Malaysia. International Journal of Wildland Fire, vol 24, pp 179–189, March 2015.
REFERENCES [18] Zaini, N.A., Noor, S.F.M, Wook, T.S.M.T, “Evaluation of APi Interface
Design By Applying Cognitive Walkthrough,” International Journal of
[1] Azman, I. & Mohd Ridwan, A. R., “Performance-based reward Advanced Computer Science and Applications, vol 10, 2019.
administration as an antecedent of job satisfaction: A case study of
Malaysia ’ s fire and rescue agencies,” Malaysian Journal of Society and [19] Zainab, H, “Study of Touch Gesture Performance by Four and Five
Space, vol 7, pp. 107-118, 2016. Year-Old Children: Point-and-Touch, Drag- and-Drop, Zoom-in and
Zoom-out, and Rotate,” Information Tech. thesis, Minnesota State
[2] Marrion, C. E., “More effectively addressing fire/disaster challenges to University, Mankato, July 2014.
protect our cultural heritage,” Journal of Cultural Heritage, vol 20, pp.
746–749, July 2016. [20] Anthony, L., Brown, Q., Nias, J., Tate, B., Mohan and S., “Interaction
and recognition challenges in interpreting children’s touch and gesture
[3] Noorhidawati, a., Ghalebandi, S. G., and Siti Hajar, R, “How Do Young input on mobile devices,” in Proceeding of the 2012 ACM International
Children Engage with Mobile Apps? Cognitive, Psychomotor, and Conference on Interactive Tabletops and Surfaces – ITS’12, 2012, p.
Affective Perspective,” Journal of Convergence Information Technology, 225.
vol 87, pp. 385-395, July. 2015.
[21] Sung, H.-Y., & Hwang, G.-J., “A Collaborative Game-based Learning
[4] Wei, W.J., & Lee, L.C., “Interactive technology for creativity in early Approach to Improving Students’ Learning Performance in Science
childhood education,” Jurnal Teknologi, vol 75(3), pp 121-126, Nov. Courses,” Computers & Education, vol 63, pp. 43–51, Nov 2012.
2015.
[22] Shi, Y.-R., & Shih, J.-L., “Game Factors and Game-Based Learning
[5] Morrongiello, B. a., Schwebel, D. C., Bell, M., Stewart, J., & Davis, A. Design Model,” International Journal of Computer Games Technology,
L., “An evaluation of The Great Escape: Can an interactive computer pp 1–11, Aug 2015.
game improve young children’s fire safety knowledge and behaviors?,”
Health Psychology, vol 31, pp 496–502, Apr. 2012. [23] Preece, J., Sharp, H., Rogers, Y., Interaction Design: Beyond Human-
Computer Interaction, vol 4, United Kingdom: Wiley, 2015.
[6] He, Q., Hong, X., Zhao, G., & Huang, X., “An Immersive Fire Training
System Using Kinect,” in Proceedings of the 2014 ACM International [24] Zaini, N.A., Noor, S.F.M, Wook, T.S.M.T, “The User Requirements of
Joint Conference on Pervasive and Ubiquitous Computing: Adjunct Game-Based Learning in Fire Safety for Preschool Children,” Journal of
Publication, 2014, paper 14, p. 231–234. Advanced Science Letters, vol 24, pp. 7795-7799, Oct 2018.
[7] Chin, L. C., & Effandi Zakaria, “Development and Validation of the [25] Singh, D. K. A., Ab Rahman, N. N. A. A., Rajikan, R., Zainudin, A.,
Game-Based Learning Module to Enhance Mathematics Achievement, Mohd Nordin, N. A., Karim, Z. A., & Yee, Y. H., “Balance and motor
Positive Learning Behaviours and Pro Social Behaviours,” Journal of skills among preschool children aged 3 to 4 years old,” Malaysian
Science And Mathematics Letters, vol 2, pp. 23–31, Jan 2014. Journal of Medicine and Health Sciences, vol 11, pp. 63–68, Jan 2015.
[8] Tang, S., Hanneghan, M., & El Rhalibi, A., Introduction to games-based [26] Oh, S. J., Fritz, M., & Schiele, B., “Adversarial Image Perturbation for
learning, Games Based Learning Advancements for Multi-Sensory Privacy Protection A Game Theory Perspective,” in Proceedings of the
Human Computer Interfaces, New York: IGI Global, 2009. IEEE International Conference on Computer Vision, 2017, p. 1491–
1500.
[9] Green, C. S., Kattner, F., Eichenbaum, A., Bediou, B., Adams, D. M.,
Mayer, R. E., & Bavelier, D., “Playing Some Video Games but Not [27] Wang, K.-C. ., Shih, S.-Y. ., Chan, W.-S. ., Wang, W.-C. ., Wang, S.-H.,
Others Is Related to Cognitive Abilities: A Critique of Unsworth et al. Gansonre, A.-A., Yeh, M.-F., “Application of building information
(2015),” Psychological Science, vol 28, pp. 679–682, 2017. modeling in designing fire evacuation-a case study,” in 31st
International Symposium on Automation and Robotics in Construction
[10] Williams-Bell, F. M., Kapralos, B., Hogue, A., Murphy, B. M., & and Mining, ISARC 2014 - Proceedings, (Isarc), 2014, p. 593–601.
Weckman, E. J., “Using Serious Games and Virtual Simulation for
Training in the Fire Service: A Review,” Fire Technology, vol 51, pp. [28] Wook, T. S. M. T., Mohamed, H., Judi, H. M., and Ashaari, N. S.,
553–584, March 2015. “Applying cognitive walkthrough to evaluate the design of SPIN
interface,” Journal of Convergence Information Technology, vol 7, pp.
106-115, March 2012.
175 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In human to human communication, context to be the identity of the user, nearby users, location, time and
increases the ability to convey ideas. However, in human to season. Other studies that discussed what context could be can
application and application to application communication, this be found in [4] [5] [6].
property is difficult to attain. Context-awareness becomes an
emergent need to achieve the goal of delivering more user-centric In 2001, Dey [7] introduced the most well-known
personalized services, especially in ubiquitous environments. definition of context: "Any information that can be used to
However, there is no agreed-upon generic framework that can be characterize the situation of an entity. An entity is a person,
reused by deployed applications to support context-awareness. In place, or object that is considered relevant to the interaction
this paper, a defeasible logic-based framework for context- between a user and an application, including the user and
awareness is proposed that can enhance the functionality of any applications themselves". This definition was a milestone in the
deployed application. The nonmonotonic nature of defeasible growth of the notion of context as it is generic, operational and
logic has the capability of attaining justifiable decisions in exceeded the boundary of (time, location and user's identity)
dynamic environments. Classical defeasible logic is extended by where context was always defined accordingly. On the other
meta-rules to increase its expressiveness power, facilitate its hand, Context-Awareness is considered to be the ability of the
representation of complex multi-context systems, and permit system to sense (gather information) about its surrounding
distributed reasoning. The framework is able to produce justified physical and operational environment at any given time,
decisions depending on both the basic functionality of the system
perceive and adapt behavior accordingly [8].
that is itself promoted by contextual knowledge and any cross-
cutting concerns that might be added by different authorities or A context-aware system should support mechanisms for
due to further improvements to the system. Active concerns that collecting contextual information, representation, reasoning
are triggered at certain contexts are encapsulated in separate and application [9]. Contextual information is domain-
defeasible theories. A proof theory is defined along with a study dependent, it can be any piece of information that describes the
of its formal properties. The framework is applied to a entity involved in the interaction, it could be time, location,
motivating scenario to approve its feasibility and the conclusions task, identity, etc. or a group of them. The acquisition of this
are analyzed using argumentation as an approach of reasoning. information is beyond the scope of this work, it is achieved
using different technologies. The emphasis of this work is on
Keywords—Context-awareness; nonmonotonicity; defeasible
logic; distributed reasoning; argumentation
the two most important phases in any framework that supports
context awareness: representation and reasoning.
I. INTRODUCTION In this paper, a generic framework is present that can guide
It is fair to say that the ubiquitous computing paradigm the contextualizing process of deployed applications. The
revolutionized our understanding of computing and what it can framework provides a powerful mechanism to represent multi-
deliver. It merges computer devices and sensors in an context distributed systems and permits distributed reasoning.
integrated environment, to provide better communication and An extension to defeasible logic theory was proposed by
enhanced accessibility to information sources. The final adding the notion of meta-rules that are able to reason over
objective is to provide users with services available whenever, theories; this enhancement would open the door of new usage
however, and wherever needed [1]. Applications should be of DL in the representation and reasoning of complex systems.
intelligent enough to handle the mobility of users and resources
themselves as well as the ever-changing context in a seamless The significance of the study lies in its conceptual analysis
manner with minimum human intervention. In other words, of context by considering it to be both, information that can
applications should be context-aware. characterize entities and information that has the ability to
characterize a whole new behavior of the system.
The term Context-Aware Computing was first introduced
in 1994 [2], this study focused on the communication aspects Another advancement of the framework is that it permits
related to broadcasting information from a server to its clients. distributed reasoning which is a challenging area in AI, as there
Context was considered to be the information related to the is no central authority to control the context flow in the overall
location of users and other objects in the system and how this system, but rather each component in the system is allowed to
information changes over time, in addition to the add its own view of manipulating contextual knowledge. This
communication overload. In [3], context awareness role in is achieved using a separation of concerns principle and can
mobile computing was discussed, the study considered context highly increase users' and administrators' satisfaction.
176 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The work is of both theoretical and empirical significance a role in deciding the placement of certain assets in the
to the research in context awareness and contextual reasoning. room e.g. a server, or turning on the air conditioning
The theoretical importance lies in the proposed extension to the which is, in turn, affects the energy consumption, and
defeasible theory that permits the representation of complex so on.
multi-context systems and facilitates distributed reasoning,
while empirical significance lies in the ability to employ the These characteristics lead to challenges that cannot be
framework to contextualize any kind of application. It allows avoided especially in complex systems that operate in
the developers of context-aware applications to easily represent ubiquitous environments where the system contains multiple
and manage different behaviors of the application in different entities and the process integration spans organizations where
contexts. interactive entities in the system may belong to different
authorities and each works under different regulations. The
This paper is organized as follows: Section 2 highlights system should be able to reason and reach justifiable decisions
some issues in contextual reasoning. Section 3 presents related regardless of these complications.
work. Section 4 presents the defeasible logic. An illustrative
scenario is presented in Section 5. Section 6 discusses our To handle these issues, a solid representation mechanism
interpretation of context and context-awareness. Section 7 should be employed that can deal with ambiguity and a
presents the proposed framework of context-awareness. concrete conflict resolution mechanism that enables inferring
Section 8 defines the formal proofs of the framework. An justifiable non-conflicted decisions. McCarthy [20] was one of
implementation of the illustrative case study in the proposed the first scientists that point out the issue of contextual
framework is presented in Section 9 along with its analysis. A reasoning. He suggested that the combination of nonmonotonic
brief discussion is presented in Section 10 and finally, reasoning and context formalism would constitute an adequate
Section 11 covers the conclusions and future work. solution to overcome the problems associated with including
contextual information in the decision-making process.
II. SOME ISSUES IN CONTEXTUAL REASONING Nonmontonicity provides mechanisms that allow the system to
reason and reach justifiable decisions by retracting conclusions
There are many alternatives in the literature that deal with that turned out to be incorrect and derive new, better-justified
knowledge representation and reasoning issues [10] [11] [12] conclusions instead [21]. This makes it very suitable to tackle
[13] [15] [14] [16] [17] [18]. However, when this knowledge is the reasoning process in dynamic situations with
characterized as contextual knowledge (i.e. "as information that incomplete/changing information.
can be used to characterize the situation of an entity"), there are
extra properties that need special treatment. Defeasible logic (DL) [22] is a well-known skeptical
nonmonotonic logic that can be used in dynamic environments
First of all, context is domain-dependent (e.g. the due to its characteristics: it is expressive, natural, not
identity of a user plays a subtle role in an access control ambiguous and programmable. It has attracted many
system, but it is not important in a supermarket billing researchers to incorporate it in different application domains
system). This is considered an appealing property that such as modeling of contracts [23], legal reasoning [24],
helps to develop personalized services. modeling social agents [25], modeling social commitments
Second, context is a conflict-sensitive concept, i.e. [15] [17] [18], etc. The most significant feature of DL is that it
multiple sources of contextual information might lead preserves the consistency of the system regardless of conflicts
to infer conflicting decisions. This happens due to because it does not produce contradictory conclusions. When a
multiple sources of contextual information which lead conflict occurs, conflicting rules do not arouse. It supports the
to ambiguity. The study in [19] highlighted other use of priorities to resolve these conflicts to allow the system
problems related to contextual information in that they inferring with incomplete/partial information.
might be unknown, imprecise, and erroneous.
III. RELATED STUDIES
Third, when reasoning is employed, context becomes There are many attempts in the literature to formalize
nested. In complex systems, the context of an entity is context in order to be able to reason based on its attributes
not merely restricted to basic contextual attributes that along with its accompanied obstacles that might lead to
are collected directly from sensors (e.g. the temperature conflicts in the decision-making process.
of a room) but rather, it refers to complex contextual
attributes that are inferred from basic contextual As the issue of context sensing and integration in highly
attributes. For example, if the temperature of the room connected to the technical infrastructure of the system, most of
is between (72 F and 76 F), the room warmth is the researches that aimed to define generic frameworks for
comfortable), in this way, a room with a temperature context awareness, pointed out the architecture aspects of the
degree (74 F) is characterized by two contextual framework, e.g. the authors in [26] proposed a context
attributes, its temperature is (74 F) and its warmth is management framework that enables the collaboration of
comfortable. This different level of abstraction gives multiple domains by exchanging contextual information. Their
context an operational power, such that a basic framework highlighted the architectural issues; it is based on a
contextual attribute may lead to a whole new behavior peer-to-peer architecture. The framework imposes a hierarchic
and direct the characterization of many other aspects in ordering of context sources and multiple reasoning tools. This
the system e.g. a room's temperature may affect not facilitates adaptability as new context sources and reasoning
only the degree of relief in the room but rather may play techniques can be added. The most important parts of the
177 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
framework are the uniform interface where all of the context- to be raw agent's knowledge received from contributed sensing
provides are attached to a reasoner where all the reasoning devices, a conceptual view of context is adopted, it considered
methods can be employed. it as a concern/goal that needs to be achieved, it is different
than the models in the literature as it defines the boundary
Other studies presented techniques to deal with contextual between what local knowledge the system is already designed
information, e.g. [27] defined a Context Toolkit that provided to manipulate (i.e. what is the input information that system
an infrastructure for prototyping context-aware applications. rules make decisions accordingly) and what is contextual
However, it didn’t provide a mechanism to reason about knowledge that could be used to enhance the system operation.
contexts. There is no formal tool to write reasoning rules for We argue that the integration of contextual information in the
contexts or to infer higher-level contexts decisions. reasoning process of a system that is driven by many concerns
Formal representation of context can be found in [28], can not only be achieved by adding additional
where an architecture and programming framework for attributes/predicates that describe contextual information and
triggering application adaptation to changes in context was additional rules that manipulate them. The projection of
proposed. It employed basic (if-then rules) to formalize the contextual knowledge on the system affects both the nature of
behavior of an application in different contexts. In [29], first- its base functionality (base concern) and the way it handles
order logic was used to describe contextual information and cross-cutting objectives, concerns or exceptions. This simulates
reasoning was done using Boolean operators and existential how humans think. Humans' decisions are never static; they are
and universal quantifiers. always changing based on upcoming knowledge, i.e. current
context. For example, a student might choose an academic
Recent trends in context-awareness pointed out the major based on his/her interest (a basic aspect), in addition to
significance of generic frameworks in manipulating context the GPA, budget, family opinion, the need for the labor market
flow in smart environments. at that time (a contextual aspect).
A formal representation of context can be found in [30] The framework is implemented using Defeasible Logic DL,
where the authors used ontologies to model information it benefits from both the expressiveness power of logic in
gathered from IoT devices in a smart home environment and representing knowledge and the nonmonotonic feature of the
used Description logic [31] to deduce activities depending on defeasible theory that facilitates a smooth reasoning process in
the gathered contextual attributes from the devices. a dynamic environment.
Another study [32] proposed a context-aware framework Based on this representation of context, the framework can
for multi-agent environment. Agents in their framework extract be viewed as a platform that can be used to augment ubiquitous
contextual information from ontologies; in fact, an agent can applications with context awareness by employing a conceptual
extract its rules and facts from one or more ontologies. Each view of context that is able to infer high-level decisions. The
agent performs reasoning based on the collected information framework allows easy integration of different modes of
and communicates with other agent(s) using bridge rules; the operations triggered by different contexts and at the same time
concluded decision is used to adapt the system behavior. The preserves the consistency of the decisions made by the system.
framework is used to generate preference sets for users, which
is a set of active rules for each user. IV. DEFEASIBLE LOGIC
Defeasible Logic DL [16] [22] had approved to be one of Defeasible logic (DL) was proposed by Nute in 2001 [22],
the famous logic tools that are successful to characterize unlike monotonic reasoning, it has a nonmonotonic relation
contextual reasoning; it has a nonmonotonic relation between between the premises and their logical consequences which
the premises and their consequences which is an effective way made it suitable for reasoning in dynamic environments. In
of formalizing the dynamic nature of ubiquities computing. order to illustrate the nonmonotonic reasoning power, assume
Several studies succeeded to build models that could reason in the situation of the following example that resembles the
the shade of contextual information based on DL [19] [33] monotonic kind of reasoning.
[34]. However, these studies handle context in an environment
Example 1: Bob is often invited to social events by his friends.
of operating agents, they consider context to be whatever local
knowledge the agent has. This view is correct and it serves the He usually attends these events; however, he has the following
goal of showing how collaborating agents can cooperate to two preferences about going to a party.
achieve a specific goal regardless of the challenges caused by P1: If the inviting person is one of his closest friends, he
the imperfect nature of context. would go.
These approaches can be viewed as enhanced versions of P2: He prefers not to go if Adam is invited.
previous approaches that aim at solving the partial knowledge
issues of autonomous agents by collaboration. This is achieved Bob was invited by his best friend, Julie, and she told him
using bridge rules [34] and mapping rules [33]. None of these that Adam is invited as well.
studies investigated the effect of context on the decision made
by each agent/entity and how contextual information can affect In a monotonic kind of reasoning, the two rules are applied
the overall behavior of the system. and both of their consequences are valid (go and don’t go)
which leads to inconsistency, it is the system developer's
The proposed framework discusses how to enhance responsibility to design rules that avoid such conflicts.
deployed applications using context. Rather than considering it Monotonic reasoning needs a lot of administrative effort and it
178 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
neither scales well nor can be used in an environment with weather is very hot or not. But, if the superiority
multiple administrative authorities. On the other hand, a relationship (R3 > R2) is introduced, then R3 overrides R2
nonmonotonic reasoning approach is founded on the ability to and it can be concluded that the weather is not very hot
infer tentative conclusions that can be retracted based on new while it is raining even during the summer season. The
evidence [14]. superiority relation > is acyclic, that is, the transitive
Formally, DL can be seen as an extension to first-order closure of > is irreflexive.
predicate calculus FOPC [35], with the addition of the The interaction of these three components permits the
defeasible implication (⇒) that is used to infer the tentative conclusion of justifiable decisions. This is referred to in Fig. 1.
conclusions, and the ambiguity-blocking priority relation (>)
that is used to preserve the consistency of the system and infer Input
justifiable conclusions in both static and dynamic domains. Justifiable
Strict Decision
Rules
Basically, a defeasible theory D (also called a knowledge
base in DL) is a triple (F, R, >), it consists of three main Defeasible
components: Facts Rules
>
179 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
personalize the functionality of the system: a location detection authority. Such that the decision made according to the urgent
service of the lecturer, schedule and the number of students invitations concern would be supported.
engaged with the professor in the online session. A system user
can set his/her own preferences based on the three available This simple scenario clarifies the challenges of using
contextual attributes (location, schedule, and status that could contextual knowledge in the decision-making process. In
be either busy or not busy depending on the number of students addition to the challenges of distributed reasoning in systems
that are active during the online session). Preferences are that encompass multiple authorities, each has its own
activated upon turning on a flag of interest on the user's mobile preferences/regulations and its own interpretation of internal
phone. For example, Professor Sami has the following rules: contextual knowledge. Each authority aims at making the
decision referring to its own rules. This motivates the need to
If he is located inside Samsung-Lab, the phone employ a distributed reasoning mechanism that can handle the
wouldn’t ring. production of justified and solid decisions in multi-context
ubiquitous environments.
If there is a scheduled lecture, the phone wouldn’t ring.
VI. CONTEXT AND CONTEXT AWARENESS
If he is engaged with less than five students in an online
session, he is not busy and the phone could ring. This Due to the enormous improvement of how computers,
rule overrules the first two rules. diffused sensors, and other devices collect situational/
contextual information, a lot of researchers tried to define
Suppose the situation when the dean asked the secretary context in several manners. Basically, context is identified by
Linda to call Professor Sami. Linda's number is in the urgent its attributes i.e. contextual information/variables that:
list; according to the anti-disturbance system rules, the phone (1) describe the user in an interaction with an application, the
should ring. However, Sami is inside Samsung Lab and is in an application/process, the environment and the interaction itself,
active session with five students, the phone should not ring. (2) can be used to deliver more user-centric personalized
From Linda's point of view, the phone should ring. She is services. The range of this information is quite vast and it
sure that her number is already listed in the urgent list; depends on the domain itself, it could be time, location,
however, she is not aware of Sami's preferences. The system is number of users, the identity of the user, user's emotional
not able to decide which argument to support, the anti- states, the focus of attention, etc. [8] [9] [36] [37].
disturbance concern argument or the users' preferences In order to build a framework that is able to enhance the
argument. Thus, an inter-concern conflict resolution operation of any application using contextual knowledge, it is
mechanism is used to regulate the decision-making process. very important to define the system's manipulated knowledge
As the end goal is to deliver personalized context-aware and enhancing contextual knowledge. Thus, throughout this
service, the designer sets that the decision inferred by users' work Dey's definition of context is extended to best suit this
preferences overrules the base system decision. In this purpose: "For a deployed application, context is any
arrangement, Sami won't be informed about the call. information used to characterize the situation of an entity and
can be sensed, collected and represented. This information is
One of the stakeholders, namely, the dean, was not satisfied not part of the group of information that already describes that
with the services provided by this system, as his secretary uses entity in the deployed application. An entity is a person, place,
the schedule of all professors and the administrative staff to or object that is considered relevant to the interaction between
determine the time of urgent meetings and she calls them based a user and an application, including the user and applications
on this knowledge. However, according to the above settings, themselves".
even though the user has no scheduled lecture at the time of the
call but is inside the lab giving advice to some students on an In this work, the set of contextual information that
online session, he/she was not informed of urgent meetings. represents domain knowledge C in an environment would be
classified according to its presence in the system, as shown in
To resolve this issue, the system should address Fig. 2:
stakeholders' concerns such as urgent invitations. The system is
connected to a meeting database that is controlled by several 1) Information that is collected from the environment in
stakeholders, it saves the time, location, invitees of meetings, the digital form or can be presented digitally, collected context,
some of them are saved in prior e.g. a workshop and some of (Co C) e.g. identity of the user, light, sound, location, size,
them are set up in an ad-hoc manner e.g. urgent meeting to etc.
discuss exams results. This concern manages the system as 2) Information that the system is designed to manipulate
follows: (Cu Co) e.g. in an access control model, the identity of the
If a person is invited, he/she should be informed. user and his role is used to determine what object(s) he/she can
access.
However, if the invitee has a scheduled lecture, he/she 3) Contextual information that can be added to enhance the
should not be informed.
functionality of the system (Ch Co) e.g. in an access control
The inter-concern conflict resolution mechanism should be model in a dynamic environment, the time and location of the
carefully designed to represent the directions of the user requesting access is of major importance.
stakeholders as they represent a higher administrative
180 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
181 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
using meta-rules. Meta-rules are rules that consequences are certain threshold. At the same time, the system is affected by
rules; they have been used in the literature as a powerful an asset safety concern, the IT department controls the air
machine that facilitates reasoning about rules for conditioning system operation regarding the safety of certain
contextualizing the provability of goals [34]. In the proposed assets in the building e.g. servers. On the other hand, the
framework their use would be extended; meta-rules are rules operation of the system is further influenced by the
that consequences are defeasible theories. maintenance department rules that turn off ACs in case of any
problems related to the hardware parts of the AC, etc.
Each trigger activates one concern using a defeasible meta-
rule such that Concerns are used to alter the behavior of the base system
by applying their conclusion. It should be noted that concerns
G ={g1,g2,….,gn} where, n≥0 is the number crosscutting do not only affect the base system, but rather affect other
concerns in the system concerns of the system; for example, the user's preferences in
Mi :gi ⇒ Di the illustrative scenario.
When a meta-rule contains an empty bode i.e. no Concerns are represented as a set of distributed defeasible
antecedent, it denotes the activation of the base system, theories D. Each theory has a unique name. System
components are referred to as,
M0 :⇒ β
Sys-c = {β} ⋃ D, where D = {D1,…, Dn}, n is the number
For the illustrative scenario, the trigger that activates the of concerns in L
preferences concern is the flag on the user's mobile phone.
The formal definition of each theory in D flows naturally
B. The Base System β from the definition of classical defeasible theory, however, the
The base system is the actual deployed application that is components of each theory would be superscripted with the
governed by rules that reflect obligations; these rules are put at concern name, e.g. concern theory Di is a tuple (FDi, RDi, >Di).
the design phase to achieve a certain purpose or goal. In this Each concern is activated by one trigger using a meta-rule.
framework β is represented using defeasible theory, it contains It should be mentioned that throughout the work of this
rules that reason about local attributes of the system (Cu) in paper, the decision inferred by β is called a base conclusion,
order to serve a certain goal. When the need arises to integrate while the decision inferred by any concern theory is
a new contextual knowledge in the decision-making process, superscripted with the name of the concern, e.g. PassD1(X),
the designer has two options, (1) If the newly added contextual means that according to concern D1, the conclusion Pass(X) is
knowledge is a simple attribute that does not crosscut the base inferred.
concern of the system and is issued by the same administrative
authority, it can be added implicitly to the base system, either D. Inter-Concerns Conflict Resolution λ
as a new rule or as a predicate in an existing rule. (2) However, Basically, concerns conclusions overrule the base system
if the newly added contextual knowledge serves a concern that conclusion. In other words, when a query is issued for a service
crosscuts the base system or is issued by a different provided by a system that includes multiple concerns, if any of
administrative authority, it will be encapsulated as a distinct these concerns concluded a decision that contradicts the
concern that is formalized. conclusion concluded by the base system, the concerns
Formally, the base system is a defeasible theory denoted as conclusion would be preferred; this is exactly where the effect
β(Fβ, Rβ, >β), The formal definition of β flows naturally from of context in changing the behavior of the system, is captured.
the definition of classical defeasible theory, however, the However, in certain contexts several concerns can be
components of the base system theory would be superscripted activated; this might lead to conflicts in the decision-making
with the base system name β. process. This case happens when the conclusion inferred from
C. Distributed Contextual Concerns Theories D one concern i.e. defeasible theory contradicts the conclusion
inferred from another concern(s). In this case, the system
Based on the separation of concerns principle, when the would use λ, a conflict resolution mechanism that follows a
collected contextual knowledge refers to a cross-cutting prioritized ordering scheme to resolve inter-concern conflicts.
concern or is issued by a different administrative authority, it
will be encapsulated in a distinct theory(s). This would λ = {(Di, Dj) ∈ Sys-c2 | (Di⊐ Dj) Di,Dj ∈ D and
considerably enhance the development, maintenance, and
security of the overall system and can enable reaching (Dk ⊐ β) ∀k Dk ∈ D}
justifiable decisions even if only partial knowledge is available. λ is a total ordering relation that is defined over system
A concern refers to the context of participating components, it uses the operator ⊐ to denote priority, such that
entities/authorities regarding the service provided by the base Di⊐ Dj states that the conclusion of Di is preferred over the
system; it reflects their interpretation of the service based on conclusion of Dj, and so on. However, it has another property;
their own manipulating of internal contextual knowledge that the definition also implies that the conclusion of any concern is
they can access. For example, suppose an energy-saving preferred over the base conclusion.
software to control an air conditioning system in a building, its It is important to notice that a total ordering relation is used
base/main concern is to manage energy consumption; it turns instead of a partial ordering relation to prioritized concerns.
ACs off for uninhabited areas when the energy level exceeds a Whenever a new concern is added, λ should be re-evaluated;
182 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
and the relation between the newly added concern and all other
system components should be set in a proper way. It is the B. System-Level Proof
designer's responsibility to decide how to prioritize concerns For a conclusion to be inferred from the framework it
based on the criticality level in the decision-making process. should be either strictly or defeasibly approved by the base
system (when no concerns are addressed) or by a higher
VIII. FORMAL PROOFS priority concern. Two types of tagged literals are introduced to
The provability of the framework would be discussed approve/not approve a conclusion:
according to the concern-level local distributed theory and the +θ B, globally approved in system L, which means that
system level theory. there is a reasoning chain that strictly or defeasibly
A. Concern Level Proof approves B in concern Di that is not defeated by any
applicable reasoning chain of a higher priority concern
Each concern is represented as a defeasible theory D, the
Dj, where both Di and Dj ∈ Sys-c.
probability of a defeasible logic is based on the concept of a
derivation (or proof) from the theory [22]. A derivation is a -θ B, not globally approved in system L, which means
finite sequence Pn=(P(1), . . ., P(n)) of tagged literals satisfying that every reasoning chain that strictly or defeasibly
the following four conditions (i.e. the inference rules for each approves B in concern Di is defeated by an applicable
of the four kinds of conclusion). reasoning chain of a higher priority concern Dj, where
Let P(1..i) denote the initial part of the sequence P n of both Di and Dj ∈ Sys-c.
length i where i ≤ n. Then a conclusion, proved subsequently The tagged literals can be formally defined by the
[16], could be either: following proof conditions:
(1) Definitely provable in D. Globally defeasibly provable in L:
+: If P(i+1) = +B then
+θ: If P(i + 1)= +θ B then
(1) B ∈ F or
(1) ((+Bβ) or (+δ Bβ)) and (D = {})) or
(2) (∃R1 ∈ Rs[B])(∀A∈Ant(R1): +A ∈P(1..i)).
(2) Not definitely provable in D. (2) (∃Mi∈ M[+δ Di] (∀gi ∈ Ant(Mi): +gi ∈ P(1..i)) and
-: If P(i+1) = -B then
(∀R1 ∈ Rs[B]) (∃A∈Ant(R1): -A ∈P(1..i)). (∃Mj∈ M[+δ Dj] (∀gj ∈ Ant(Mj): +gj ∈ P(1..i)) and
183 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
β = (Fβ, Rβ, >β), Definition: Suppose D (F, R, >), is a defeasible logic theory
β and B is a ground literal. Arg is said to be an argument that
F = {calling(X,Y), unknown(X), international(X)} supports the conclusion B from D, denoted by <Arg, B>, if Arg
R1β :calling(X,Y) in-urgent(X) ⇒ ring(Y) is a minimal set of defeasible rules (Arg ⊆ Rd), such that:
R2β :calling(X,Y) international(X)⇒ ring(Y) 1) B can be defeasibly derivate from (Arg ∪ F ∪Rs),
2) No pair of contradictory literals can be defeasibly
R3β :calling(X,Y) unknown ⇒ ¬ring(Y) derived from (Arg ∪ F ∪Rs), and
>β = {( R1β > R3β), (R2β > R3β) } 3) Arg contains no rule that contains an antecedent that is
complementary to an antecedent of another rule in Arg.
However, Concern D1 encodes the lecturer's preferences
regarding call management. The lecturer makes his decision With respect to analyzing the behavior of the theory in the
based on three contextual attributes, his location, schedule and case study using argumentation, suppose the situation when the
his status that could be either busy or not based on the number dean asked the secretary Linda to call Professor Sami.
of students that are active during the online session. Professor Professor Sami has the preferences flag set on. Linda's number
Sami's preferences are formalized by a contextual concern is in the urgent list; Sami is inside Samsung Lab and is in an
theory D1 which is activated using meta-rule M1 due to a flag active session with five students.
on the user's phone. There are two arguments that support conflicting
M1: flag(on) ⇒ D1 conclusions:
184 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
distributed systems. It complies with the main characteristics of framework. We believe the management of this issue can
ubiquitous and distributed systems in providing transparency, present a magnificent step in the field of context-awareness.
reliability, and scalability. At the same time, it enables the REFERENCES
evaluation of distributed decisions and produces globally
[1] Park, I. S., Kim, W. T., & Park, Y. J. (2004, February). A ubiquitous
justifiable conclusions. This is achieved using triggers and the streaming framework for multimedia broadcasting services with QoS
relation between active concerns, unlike the bridge rules and based mobility support. In International Conference on Information
mapping rules that were used in literature. Networking (pp. 65-74). Springer Berlin Heidelberg.
[2] Schilit, B. N., & Theimer, M. M. (1994). Disseminating active map
The consistency of the system is attained by the consistency information to mobile hosts. IEEE network, 8(5), 22-32.
of defeasible logic itself as for any statement; there is a [3] Brown, P. J., Bovey, J. D., & Chen, X. (1997). Context-aware
proof/reasoning chain that can determine whether or not that applications: from the laboratory to the marketplace. IEEE personal
statement holds and inconsistencies can be detected using the communications, 4(5), 58-64.
proof theory. [4] Capurso, N., Mei, B., Song, T., Cheng, X., & Yu, J. (2018). A survey on
key fields of context awareness for mobile devices. Journal of Network
XI. CONCLUSIONS AND FUTURE WORK and Computer Applications, 118, 44-60.
[5] Schmidt, A., Beigl, M., & Gellersen, H. W. (1999). There is more to
In this paper, we proposed a novel framework for context- context than location. Computers & Graphics, 23(6), 893-901.
awareness that can contextualize any deployed application. The [6] Gruber, T. R., Brigham, C. D., Keen, D. S., Novick, G., & Phipps, B. S.
framework is based on a conceptual analysis of context; it (2018). U.S. Patent No. 9,858,925. Washington, DC: U.S. Patent and
captures the behavior of contextual knowledge as it penetrates Trademark Office.
into deployed applications. It fairly simulates how the human [7] Dey, A. K. (2001). Understanding and using context. Personal and
being perceives context either as plain attributes or as concerns ubiquitous computing, 5(1), 4-7.
that need to be considered in order to make better decisions, [8] Fischer, G. (2012). Context-aware systems: the 'right' information, at the
change behavior and personalize services. 'right' time, in the 'right' place, in the 'right' way, to the 'right' person. In
Proceedings of the International Working Conference on Advanced
The framework is efficiently mapped to defeasible theory. Visual Interfaces (pp. 287-294). ACM.
It is generic, flexible and scalable. It allows the system to make [9] Ryan, N., Pascoe, J., & Morse, D. (1999). Enhanced reality fieldwork:
justifiable decisions regardless of the number of available the context aware archaeological assistant. Bar International Series, 750,
contextual attributes, concerns, or the number of administrative 269-274.
authorities that control the decision-making process. Its main [10] Pollock, J. L. (1996). OSCAR: A general-purpose defeasible
reasoner. Journal of applied non-classical logics, 6(1), 89-113.
strength lies in its distributed approach of reasoning and its
[11] Moubaiddin, A., & Obeid, N. (2009). Partial information basis for agent-
ability to represent concerns in defeasible theories. based collaborative dialogue. Applied Intelligence, 30(2), 142-167.
The analysis showed that the framework is able to capture [12] Obeid, N., & Moubaiddin, A. (2010). Towards a formal model of
both the contextual aspects and the concerns of different knowledge sharing in complex systems. In Smart Information and
Knowledge Management (pp. 53-82). Springer, Berlin, Heidelberg.
authorities in the system. The consistency in the system is
[13] Obeid, N., & Rao, R. B. (2010). On integrating event definition and
attained by two levels of conflict resolution mechanisms, event detection. Knowledge and information systems, 22(2), 129-158.
concern level, and system level.
[14] Obeid, N. (2012). Three-Values Logic and Non-Monotonic
The proposed extension of the defeasible theory using Reasoning. COMPUTING AND INFORMATICS, 15(6), 509-530.
meta-rules improved the expressiveness power of the logic [15] Moubaiddin, A., & Obeid, N. (2013). On formalizing social
commitments in dialogue and argumentation models using temporal
through enabling nonmonotonic reasoning over sets of defeasible logic. Knowledge and information systems, 37(2), 417-452.
defeasible theories rather than defeasible rules.
[16] Sabri, K. E., & Obeid, N. (2016). A temporal defeasible logic for
We have investigated the capabilities of the system in handling access control policies. Applied Intelligence, 44(1), 30-42.
reasoning in environments with multiple entities that have [17] Moubaiddin, A., Salah, I., & Obeid, N. (2018). A temporal modal
cross-cutting concerns. Future work may exploit the flexibility defeasible logic for formalizing social commitments in dialogue and
argumentation models. Applied Intelligence, 48(3), 608-627.
of the proposed framework and its augmented power of
[18] Mobaiddin, A., & Obeid, N. (2018, June). On Commitments Creation,
expressing complex systems in providing personalized services Compliance and Violation. In International Conference on Industrial,
i.e. entities/users that share the same concern but each one of Engineering and Other Applications of Applied Intelligent Systems (pp.
them preserves its own right of manipulating contextual 465-476). Springer, Cham.
knowledge in its own way. For example, according to the [19] Bikakis, A. and Antoniou, G. (2010). Rule-based contextual reasoning in
scenario, more than one user has the same concern (e.g. ambient intelligence. In International Workshop on Rules and Rule
preferences) but each of them has its own setting of Markup Languages for the Semantic Web (pp. 74-88). Springer Berlin
Heidelberg.
preferences.
[20] McCarthy, J. (1987). Generality in Artificial
Further work would also consider investigating the Intelligence. Communications of the ACM, 30(12), 1030-1035.
capabilities of this framework by implementing it on real- [21] Antoniou, G., & Williams, M. A. (1997). Nonmonotonic reasoning. Mit
world ubiquitous systems where context plays an important Press.
role. [22] Nute, D. (2001, October). Defeasible logic. In International Conference
on Applications of Prolog (pp. 151-169). Springer Berlin Heidelberg.
Another aspect to be considered in contextual reasoning is [23] Governatori G (2005) Representing business contracts in RuleML.
the effect of context on the manipulation of the prioritizing International Journal of Cooperative Information Systems 14(2-3):181–
scheme of both the classical defeasible logic and the proposed 216.
185 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[24] Governatori, G., Olivieri, F., Scannapieco, S., & Cristani, M. (2012). [33] Antoniou, G., Bikakis, A., Karamolegou, A., & Papachristodoulou, N.
Revision of defeasible logic preferences. arXiv preprint (2006). A Context-Aware Meeting Alert Using Semantic Web and Rule
arXiv:1206.5833. Technology-Preliminary Report. Semantic Web Technology For
[25] Governatori G, Rotolo A, Padmanabhan V (2006) The cost of social Ubiquitous & Mobile Applications (SWUMA’06), 23.
agents. In: Proceedings of the AAMAS 2006, pp 513–520. [34] Dastani, M., Governatori, G., Rotolo, A., Song, I., & Van Der Torre, L.
[26] Van Kranenburg, H., Bargh, M. S., Iacob, S., & Peddemors, A. (2006). (2007, November). Contextual agent deliberation in defeasible logic.
A context management framework for supporting context-aware In Pacific Rim International Conference on Multi-Agents (pp. 98-109).
distributed applications. IEEE Communications Magazine, 44(8), 67-74. Springer, Berlin, Heidelberg.
[27] Dey AK et al. (2001) A conceptual framework and a toolkit for [35] Harel, D. (1979). First-order dynamic logic (Vol. 68). Berlin: Springer.
supporting the rapid prototyping of context-aware applications. Cont Aw [36] Al-Zyoud, M., Salah, I., & Obeid, N. (2012, November). Towards a
Comput-HCI J 16:97–116. model of context awareness using web services. In International
[28] Schilit WN (1995). A context-aware system architecture for mobile Conference on Computational Collective Intelligence (pp. 121-131).
distributed computing. Dissertation, Columbia University. Springer, Berlin, Heidelberg.
[29] Ranganathan, A., & Campbell, R. H. (2003). An infrastructure for [37] Musumba, G. W. and Nyongesa, H. O. (2013). Context awareness in
context-awareness based on first order logic. Personal and Ubiquitous mobile computing: A review. International Journal of Machine Learning
Computing, 7(6), 353-364. and Applications, 2(1), 5-pages.
[30] Alirezaie, M., Renoux, J., Köckemann, U., Kristoffersson, A., Karlsson, [38] Antoniou, G., Maher, M. J., & Billington, D. (2000). Defeasible logic
L., Blomqvist, E., ... & Loutfi, A. (2017). An ontology-based context- versus logic programming without negation as failure. The Journal of
aware system for smart homes: E-care@ home. Sensors, 17(7), 1586. Logic Programming, 42(1), 47-57.
[31] Obeid, M., Obeid, Z., Moubaiddin, A., & Obeid, N. (2019, July). Using [39] Moubaiddin, A., & Obeid, N. (2008). Dialogue and argumentation in
Description Logic and Abox Abduction to Capture Medical Diagnosis. multi-agent diagnosis. In New Challenges in Applied Intelligence
In International Conference on Industrial, Engineering and Other Technologies (pp. 13-22). Springer, Berlin, Heidelberg.
Applications of Applied Intelligent Systems (pp. 376-388). Springer, [40] García, A. J., & Simari, G. R. (2014). Defeasible logic programming:
Cham. Delp-servers, contextual queries, and explanations for
[32] Uddin, I., Rakib, A., Haque, H. M. U., & Vinh, P. C. (2018). Modeling answers. Argument & Computation, 5(1), 63-88.
and reasoning about preference-based context-aware agents over
heterogeneous knowledge sources. Mobile Networks and
Applications, 23(1), 13-26.
186 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Prosody refers to structure of sound and rhythm and 150.61Hz and in sound duration of 0.721ms and 0.566ms
and both are essential parts of speech processing applications. It seconds respectively can be seen in “Fig. 1” and “Fig. 2”.
comprises of tone, stress, intonation and rhythm. Pitch and
duration are the core elements of acoustic and that information During the literature review it is observed that the pitch
can make easy to design and development for application and duration for Sindhi sounds have not been digitally
module. Through these two peculiarities, the prosody module can analyzed at acceptable level and statistically measurement also
be validated. These two factors have been investigated using the have not been done for Sindhi sounds parameters like pitch
sounds of Sindhi adults and presented in this paper. For the and duration whereas deep analysis is mandatory for various
experiment and analysis, 245 male and female undergraduate speech processing application specifically prosody generation.
students were selected as speakers belonging from five different Therefore pitch and duration is statistically measured and
districts of upper Sindh and categorized into groups according to presented in this paper.
their age. Particular sentences were given and recorded
individually from the speakers. Afterward, these sentences The development of Prosody Generation Module is the
segmented into words and stored in a database consisting of 1960 main objective of this research for which the pitch and
sounds. Thus, distance of the frequency in pitch was measured duration ranges are the prerequisites parameters. In solitude
via Standard Deviation (SD). The lowest Mean SD accompanied condition of prosody, the unusual effects of prosody are
0.25Hz and 0.28Hz received from male and female group of complicated to reproduce and also the analysing of prosody is
district Sukkur. The highest Mean SD has measured with male difficult due to the function multiplicity [3]. The prosody
and female group of district Ghotki along 0.42Hz and 0.49Hz. generation module is the mandatory components of various
Generally, the pitch of female’s speakers was found high in speech processing software applications specifically business-
contrast to male’s speaker by 0.072Hz variation. related Text-to-Speech systems today make use of rather
unsophisticated systems, characteristically conveying a
Keywords—Prosody generation; speech analysis; pitch; defaulting sentence accent based on the function word
duration; Sindhi sounds distinction [4].
I. INTRODUCTION In this study, various male and female adult inhabitants of
Sindhi Language is being spoken with various accents five districts: Khairpur (K), Sukkur (S), Ghotki (G), Shikarpur
across the Sindh. Linguists generally divided this language in (Sh) and Larkana (L) are chosen for recording the sounds to
six dialects [1]. People speak the language in different accent evaluate the fundamental frequencies particularly pitch and
in the same region because either they have migrated from duration through which prosodic information pertaining to
other region or they are living in the districts adjacent to recorded sounds can be depicted and analyzed for further
Punjab and Balochistan provinces. processing and development of speech based software
applications [5] [6].
Phonologically, Sindhi language is rich and has sufficient
sound inventory [2]. The complex variation in accent is major
cause for the less accuracy in Sindhi speech processing
software applications specifically prosody generation module.
To reach the maximum accuracy in software applications it is
mandatory to measure the fundamental frequency of Sindhi
sounds.
While speaking the Sindhi language, the variations in
sound duration and pitch are normally observed with routine
sounds of words but surprisingly these variations are also
observed when the homographic words are spoken with the
different diacritic symbols even spoken by the same adult. In
Sindhi, huge number of homographic words is available which
are commonly used and spoken. For instance س ُر ُ andسرُ are
the homographic words used as singular and plural
respectively having difference in pitch frequency of 146.89Hz Fig. 1. Sound Information of Word س ُُر
ُ using PRAAT.
187 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
188 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
189 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
190 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. INVESTIGATED MEAN PITCH AND DURATION OF RECORDED SOUNDS OF 1 LETTER WORDS
TABLE. III. INVESTIGATED MEAN PITCH AND DURATION OF RECORDED SOUNDS OF 2 LETTER WORDS
191 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. IV. INVESTIGATED MEAN PITCH AND DURATION OF RECORDED SOUNDS OF 3 LETTER WORDS
TABLE. V. INVESTIGATED MEAN PITCH AND DURATION OF RECORDED SOUNDS OF 4 LETTER WORDS
192 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. VI. INVESTIGATED MEAN PITCH AND DURATION OF RECORDED SOUNDS OF 5 LETTER WORDS
TABLE. VII. INVESTIGATED MEAN PITCH AND DURATION OF RECORDED SOUNDS OF 6 LETTER WORDS
193 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The calculated results of words having 6 letters are given The sentences were given to speakers for investigation of
in Table VII. The least MDM of 0.5422ms and the highest pitch and duration from the recorded sounds. For this, 245
MDM of 0.5576ms are measured. The minimum MDF is undergraduate students are selected from the five Districts;
0.4723ms and the maximum MDF is 0.4854ms. The lowest Khairpur, Sukkur, Ghotki, Shikarpur and Larkana of Upper
MPM is 150.94Hz and the highest MPM is 151.66Hz. The Sindh having different ages. Eight sentences were given
lowest MPF of 240.29Hz and largest calculated MPF of individually which are randomly selected from the prepared
241.38Hz are received from the recorded male and female 65 sentence. The total words were spoken by male and female
recorded sounds. The calculated results of Mean SD are speakers comprises on 1960.
shown in “Fig. 9”. The mean SD of Khairpur shows that male
and female pitch with large words are same and the peoples Furthermore, all the recordings are recorded at the radio
speak slightly. The high SD results are received with the station, Khairpur on different timings and due to short time
speakers of Ghotki. only limited time was given by the authority. The recorded
sounds are segmented into words and then stored into the
computer. The PRAAT speech analyzer is used for
segmentation of sounds of words and analysis of the recorded
sounds.
The duration and pitch of the recorded sounds are
separately calculated and presented according to the words
based on the number of letters. The lowest MPM 127.76Hz is
calculated with 5 letter words and the highest MPM of
159.78Hzis recorded with 1 letter word. Experimented results
proved that the pitch in Sindhi sounds is entirely based on the
syllabification. Almost, 1 letter word has one syllable while it
is possible the number of syllables increases when word is
based on more than 1 letters. It is also observed that the pitch
of Sindhi people is high at the start of the word and at the end
of the syllable particularly when syllable ends with „Jazm‟.
Fig. 8. Mean SD of Male and Female Pitch with 5 Letter Words.
The cumulative Mean SD of all words spoken by Male and
Female speakers is depicted in “Fig. 10”.
On the basis of received results and the calculate Mean SD
of all words spoken by the male speakers it is found that
speakers pertaining to district Ghotki have high pitch as
compare to others. And the speakers of Larkana district have
low pitch. However, little variation in pitch is observed with
the male speakers of Khairpur, Sukkur and Shikarpur
Districts. Approximately same results of the female speakers
are received during the experiments. It is found that the pitch
of the female speakers are high then the pitch of male speakers
but the variation is only 0.072.
194 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
195 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—C2C (consumer-to-consumer) is a business model lot of attention from researchers, communities and industry.
where two individuals transact or conduct business with each Deep learning is able to provide excellent results for various
other using a platform. A consumer act as a seller put their tasks such as the traffic signs identification [8], mandarin letter
product in a platform later will be displayed to another consumer writing identification [9], etc.
act as a buyer. This condition encourages platform to maintain
high quality product information especially image that is Deep learning implementation from scratch requires huge
provided by the seller. Product images need to be relevant to the amount of dataset. This can be an obstacle because the huge
product itself. It can be controlled automatically using image data collection requires a lot of resource. Besides that, in
classification. In this paper, we carried out a research to find out various cases, the data needed is difficult to collect. Under
the best deep learning model in image classification for e- these conditions, deep learning can still be applied by transfer
commerce products. A dataset of 12,500 product images is learning method. Transfer learning refer to the situation where
collected from various web sources to be used in training and what has been learned in one setting is exploited to improve
testing process. Five models are selected and fine-tuned using a generalization in another setting [10]. In case of image
uniform hyperparameter set-up. Those hyperparameters are classification using deep learning, this can be done by using a
found by using a manual process by trying a lot of previously trained model that is often called the pre-trained
hyperparameters. The testing result from every model is model and fine-tuned that model using a target dataset.
presented and evaluated. The result shows that NASNetLarge
yield the best performance among all evaluated models with 84% The use of pre-trained models is very helpful because it can
testing accuracy. save time and costs of the training process. In this study several
pre-trained models will be re-trained using fine-tuning method
Keywords—Image classification; e-commerce; product images; on a 12,500 product images dataset. Then their performances
deep learning; hyperparameter tuning will be compared based on the testing accuracy. Although there
are previous studies related to image classification using deep
I. INTRODUCTION learning, it is relatively difficult to find study on the image
The current adoption of e-commerce in Indonesia is high. classification process that addresses everything from data
Das et al. [1] mentioned in 2016, 78% or more than 80 million collection to the use of several models and comparing the
users had made online purchases. The online transactions that results, specifically in the area of e-commerce.
occur in Indonesia, one of which occurs in the C2C business The rest of this paper is organized as follows. Previous
model, where two individuals seller and buyer transact with works on related topics are presented in Section II. Background
each other [2]. theories of the pre-trained models, which are used in our
In the C2C business model, there is a platform that research, are described in Section III. We describe our datasets
mediates between sellers and buyers. The seller advertises their in Section IV. Experiments and Results are presented in
products on a platform, which will then be seen by the buyers. Sections V and VI, respectively. Discussion on the
This causes weak control of the information contained in the experimental results is presented in Section VII. Finally, we
products displayed on the platform to the buyers [2]. Product conclude our paper in Section VIII.
information, such as product image is an important factor for
successful transaction. Several researches show that the image II. RELATED WORK
of the product is very important in buying interest [3], [4], [5]. There are several papers related to the importance of the
Therefore, a mechanism is needed to maintain the quality of product information’s quality on e-commerce, such as [11],
product images uploaded by the seller. An automatic approach [12] and [13]. Those papers discuss the quality of information,
using image classification method can be used to achieve it. which includes content accuracy, completeness, and relevance.
There are many algorithms that can be used for image The accuracy, which is one of the dimensions of information
classification. One algorithm that is currently popular to solve quality, represents the perception of consumers that the
image classification problems is Convolutional Neural information presented for a product or other content on the
Network (CNN) [6], [7], which is one of the deep learning platform is true [13]. One of the product information
algorithms. Deep learning itself in recent years has received a dimension that will help provide a buyer understanding of a
196 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
product is image. In the process of creating product block with stride of 2. Those two blocks are stacked to form
information, there may be a mistake that causes a decrease in MobileNetV2 as shown in Fig. 2.
the quality of product information. So to prevent this, a
mechanism to maintain the quality of information or product C. NASNetMobile
images remain relevant and accurate is needed. One way is to NASNetMobile is one of the variants of NASNet
automatically classify which images are good and which of architecture for mobile platforms. Based on the Keras version
them are not good. 2.2.4, NASNetMobile has a total of 769 layers. It has more
Regarding the image classification itself, paper [14] layers than MobileNet models. The number of layers that will
provides an overview of the process of image classification. be frozen is 724 and will use input dimension similar with
Generally, it covers image pre-processing, feature extraction, MobileNet that is 224x224. NASNet architecture consist of
and classification. Image pre-processing is needed before an several cell stacked together: normal cell and reduction cell.
image is analysed. It can be in the form of image For NASNetMobile, every 4 normal cell stacked together
normalization. The feature extraction than do the image followed by a reduction cell, as shown in Fig. 3.
transformation to understand the image. In the end, the
classification process is done to identify an image as a class D. NASNetLarge
from a group of classes. This is the largest model from NASNet. This model
Various deep learning algorithms have been used for image achieved top-1 accuracy for ImageNet at 82.7%. This
classification, such as MobileNet, NASNet, and DenseNet. performance is similar to SENet as the winner of ILSVRC
MobileNet is a small-sized model optimized for use on mobile 2017. Based on Keras library, this model has 1039 layers. The
devices [15], [16]. Although the size of the model is small but first 950 layers will be frozen. This model will use 331x331
its performance on Imagenet outperforms GoogleNet, which input size. NASNetLarge has similar architecture with
was the winner of the 2014 ILSVRC. There is also NASNet, NASNetMobile it is consist of several cells stacked together.
which managed to match the performance of SENet, which is
the winner of ILSVRC 2017 [17]. There is also DenseNet,
which performed as good as ResNet, which is the winner of
ILSVRC 2015 [18].
B. MobileNetV2
This is an improvement to MobileNetV1. For the ImageNet
dataset, this newly MobileNet architecture improves the state
of the art for wide range of performance points [16]. Using
Keras library, total layer of MobileNetV2 is 155 layers without
top layers. The first 135 layers will be frozen and will use
224x224 as input size. In MobileNetV2, there are two types of
blocks. One is residual block with stride of 1. Another one is
Fig. 2. MobileNetV2 Architecture Blocks.
197 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
E. DenseNet121
The smallest variant of DenseNet available in Keras library
is chosen. It has been shown to yield performance similar to
ResNet101, but with less parameters [18]. Based on Keras
library, it has a total of 427 layers. The first 411 layers will be
frozen and will use input size of 224x224. DenseNet121
consist of 4 dense blocks. Every dense block consists of several
convolution block. Fig. 4 shows the convolution block that is
stacked together to construct a dense block.
IV. DATASETS
A total of 12,500 product images are collected from several
C2C e-commerce websites. The collection process is done Fig. 5. Images Dataset Collection Process.
manually by visiting a web page that contains images related to
a category. For example, to collect images for trouser category,
a webpage containing all trouser images is visited and then by
using a chrome extension all images are downloaded and saved
to a local folder. This process is repeated until a category
contains desired total number of images. Fig. 5 shows the
process of collecting the product images for the dataset.
The dataset is organized into two sub-groups: 10.000
training data and 2.500 testing data. Each sub-group has 10
balance categories with 1.000 images for each category in
training dataset and 250 images for each category in testing
dataset. Validation dataset will be obtained by performing a
random split on the training dataset with split ratio of 0.2.
Fig. 6 shows the dataset distribution across all categories. Fig. 6. The Distribution of 12,500 Product Images Dataset.
198 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Data Augmentation
The image dataset will be loaded batch by batch, then for
every batch of images, data augmentation process is
performed. It is used to introduce variation and noise to the
model. The input image is rotated, shifted, shear, zoomed, and
flipped horizontally. Augmentation process will not be
performed for the testing data. Fig. 8 shows the data after
augmentation process.
C. Model Training
The training process uses transfer learning method for 5
pre-trained models and uses the Tensorflow and Keras library.
The 5 pre-trained models are available in the Keras library. The
training process is done one-by-one for each model. The
original weight of the model is used, and the classification
Fig. 7. Sample of 12,500 Product Images Dataset. layer of the original model is not included, so that it is possible
to add new classification layer that matches the 12,500 product
V. EXPERIMENTS images dataset characteristic. Training process is performed for
100 epochs using the same hyperparameter values for all 5
Training for each model is done using a uniform models.
hyperparameter set-up. Those hyperparameters is found by
using a manual process by trying a lot of hyperparameter and D. Model Testing
training a model using each hyperparameter. The number of
epoch is set to 100. It is chosen due to the fact that some After the training process is complete, it continues with the
models beginning to show overfitting at the end of 100 epoch. testing process using 2,500 testing images from the dataset.
Also, NASNetLarge requires more time for training. The result of testing accuracy from this process will be
evaluated and compared to find out which model has the best
The experiment for every model is performed according to performance.
the following steps:
1) Finding hyperparameter: This is a set of activities to
find suitable hyperparameter to train all models.
2) Data augmentation: This is performed to the training
and validation data.
3) Model training: It uses Keras’ fit_generator to feed
sample data using batch to the model.
4) Model testing: The model then tested using 2.500
testing sub-group from the dataset to find out testing accuracy.
These steps are applied similarly to all 5 models. The
differences are just in the input dimension and the total number
of frozen layers.
A. Finding Hyperparameter
Model training is done using transfer learning. A fine-
tuning process is performed for all models using the same Fig. 8. Sample Data after Augmentation Process.
199 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
This section provides the results from every experiments, Category Precision Recall F1-Score
which includes validation and testing performances. Trouser 75% 85% 80%
Short 80% 72% 76%
A. Finding Hyperparameter
Long-sleeved shirt 83% 40% 54%
The result of this process is optimal hyperparameter values
for learning rate, validation split, and also number of Short-sleeved shirt 64% 85% 73%
bottleneck layer included in the training. Shoes 92% 66% 77%
There are three values compared for every hyperparameter. Hat 88% 87% 87%
Each value is then used to train MobileNet and the most Bag 90% 84% 87%
optimal hyperparameter, based on validation accuracy, will be Sandal 74% 89% 80%
used to train another models. Learning rate are compared at
0.001, 0.0001, and 0.00001. The result shows that learning rate Jacket 67% 88% 76%
of 0.0001 gives optimal result based on validation accuracy. T-Shirt 80% 79% 79%
Validation split are evaluated at 0.20, 0.25, and 0.30. The most Average: 79% 77% 77%
optimal result is validation split ratio of 0.20. Different
numbers of bottleneck layer can be included in re-training
process are also tested. The first experiment uses 0 layer, which
means all layer from original model architecture is frozen. The
second training is started from the last 1 block, and the third
one is from the last 2 blocks. The result shows that the training
from the last 2 blocks gives the most optimal result. It yields
the highest validation accuracy after 100 epochs.
B. MobileNetV1
MobileNetV1 gives validation accuracy of 89.34% and
testing accuracy of 82%. Fig. 9 shows the confusion matrix of
the testing results for MobileNetV1.
Table I summarizes the classification report on testing
dataset for MobileNetV1. As shown in the table, it gives an
average precision of 79%, average recall of 77%, and average
f1-score of 77%.
200 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
D. NASNetMobile
NASNetMobile gives validation accuracy of 84% and
testing accuracy of 78%. Fig. 11 shows the confusion matrix of
the testing results for NASNetMobile.
Table III summarizes the classification report on testing
dataset for NASNetMobile. As shown in the table, it gives an
average precision of 79%, average recall of 78%, and average
f1-score of 78%.
E. NASNetLarge
NASNetLarge is the biggest model in term of architecture.
It gives validation accuracy of 90.69% and testing accuracy of
84%. Fig. 12 shows the confusion matrix of the testing results
for NASNetLarge.
Fig. 12. Confusion Matrix for NASNetLarge.
Table IV summarizes the classification report on testing
dataset for NASNetLarge. As shown in the table, it gives an
average precision of 84%, average recall of 84%, and average TABLE. IV. CLASSIFICATION REPORT OF NASNETLARGE
f1-score of 84%.
Category Precision Recall F1-Score
Trouser 84% 86% 85%
Short 87% 79% 83%
Long-sleeved shirt 74% 75% 75%
Short-sleeved shirt 73% 84% 78%
Shoes 90% 86% 88%
Hat 92% 89% 90%
Bag 90% 91% 90%
Sandal 84% 89% 87%
Jacket 85% 77% 81%
T-Shirt 82% 84% 83%
Average: 84% 84% 84%
F. DenseNet121
DenseNet121 is the smallest variant from DenseNet model.
It is comparable to other mobile models. DenseNet121 gives
validation accuracy of 84.09% and testing accuracy of 75%.
Fig. 11. Confusion Matrix for NASNetMobile.
Fig. 13 shows the confusion matrix of the testing results for
DenseNet121.
201 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. V. CLASSIFICATION REPORT OF DENSENET121 Fig. 15. Validation Loss for 100 Epochs.
202 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[3] X. Li, M. Wang, and Y. Chen, “the Impact of Product Photo on Online
VIII. CONCLUSIONS Consumer Purchase Intention: an Image-Processing Enabled Empirical
Study,” PACIS 2014 Proc., 2014.
In this work, e-commerce product images classification has [4] C. Lumb, “The Customer Decision Process and User Interaction in E-
been demonstrated using deep learning algorithm. We have commerce,” 2014.
collected and labelled a total of 12,500 product images dataset. [5] E. Huang and C.-C. Liu, “A Study on Trust Building and Its Derived
The images were crawled form several C2C e-commerce Value in C2C E-Commerce,” J. Glob. Bus. Manag, 2010.
websites. Five deep learning models have been evaluated on [6] A. KrizhKrizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet
the dataset, which include MobileNetV1, MobileNetV2, Classification with Deep Convolutional Neural Networks. Advances In
NASNetMobile, NASNetLarge, and DenseNet121. Neural Information Processing Systems, 1–9.evsky, I. Sutskever, and G.
E. Hinton, “ImageNet Classification with Deep Convolutional Neural
Based on the experimental results, NASNetLarge achieves Networks,” Adv. Neural Inf. Process. Syst., 2012.
the best performance for image classification with testing [7] J. Patterson and A. Gibson, Deep Learning A Practitioner’s Approach, 1st
accuracy of 84%. Also, it shows the best performance during ed. boston: oreilly, 2016.
training with validation accuracy of 90.69%. It outperforms the [8] D. Cireşan, U. Meier, and J. Schmidhuber, “A committee of neural
other four models that are trained using similar networks for traffic sign classification,” in Proceedings of the
International Joint Conference on Neural Networks, 2011.
hyperparameter. However, this performance comes with a cost
[9] D. Ciresan, U. Meier, and J. Schmidhuber, “Multi-column Deep Neural
of larger architecture and longer training time compared to the Networks for Image Classification,” Int. Conf. Pattern Recognit, 2012.
other models. [10] A. C. Ian Goodfellow, Yoshua Bengio, Deep Learning. 2016.
Further research related to this study can be done by [11] T. Singh Chhikara, “Information Quality -Crucial Aspect of E-
making variations to the dataset, using imbalanced dataset, or Commerce,” IOSR J. VLSI Signal Process. Ver. II, 2015.
increasing the number of images. Another research can also be [12] W. H. DeLone and E. R. Mclean, “The DeLone and McLean Model of
Information Systems Success: A Ten-Year Update,” J. Manag. Inf. Syst. /
done by conducting experiments using different sets of Spring, vol. 19, no. 4, pp. 9–30, 2003.
hyperparameters.
[13] B. H. Wixom and P. A. Todd, “A Theoretical Integration of User
Satisfaction and Technology Acceptance,” Info. Sys. Res., 2005.
ACKNOWLEDGMENT [14] P. Kamavisdar, S. Saluja, and S. Agrawal, “A survey on image
classification approaches and techniques,” Int. J. Adv. …, 2013.
This work was supported by the Directorate General of [15] A. G. Howard et al., “MobileNets: Efficient Convolutional Neural
Strengthening for Research and Development, Ministry of Networks for Mobile Vision Applications,” in arxiv, 2017.
Research, Technology, and Higher Education, Republic of [16] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. C. Chen,
Indonesia, through research grant number “MobileNetV2: Inverted Residuals and Linear Bottlenecks,” in
Proceedings of the IEEE Computer Society Conference on Computer
23/AKM/MONOPNT/2019. Vision and Pattern Recognition, 2018.
REFERENCES [17] B. Zoph, V. Vasudevan, J. Shlens, and Q. V. Le, “Learning Transferable
[1] K. Das, M. Gryseels, P. Sudhir, and K. T. Tan, “Unlocking Indonesia’s Architectures for Scalable Image Recognition,” in Proceedings of the
digital opportunity,” no. October, pp. 1–28, 2016. IEEE Computer Society Conference on Computer Vision and Pattern
Recognition, 2018.
[2] C. Dan, “Consumer-To-Consumer (C2C) Electronic Commerce: The
Recent Picture,” Int. J. Networks Commun, 2014. [18] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely
connected convolutional networks,” in Proceedings - 30th IEEE
Conference on Computer Vision and Pattern Recognition, CVPR 2017,
2017.
203 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This paper proposes a gamification design model concept of game-based learning which principally aims at
that can be used to design and develop gamified e-learning delivering learning content and objectives in an entertaining
systems. Furthermore, a controlled and carefully designed approach implemented as a game so that learners can play the
experimental evaluation in terms of learning effectiveness of game in order to learn.
gamification is offered. The experiment was conducted with 44
participants randomly assigned to an experimental ‘gamified’ Based on the literature and recent reviews on gamification
condition and a controlled ‘non-gamified’ condition. In both [4], [14], [15], several issues have been found. First, most
conditions the same learning material, to teach computer gamification studies were not supported with empirical
security, were used. The main difference between the two evidence following well-designed, controlled and thorough
conditions was the integration of gamification in an e-learning experimental evaluations that have control conditions. Second,
system designed based on the proposed model. The results mixed results can also be found generally with limited sample
indicate that learning using the gamified version of the e-learning sizes. Third, studies were mostly based on short-term
system produces better short-term and medium-term learning applications of some types of game elements. Fourth, some
gain than learning using the non-gamified e-learning version. studies do not rely on such models as a foundation to the
Future avenues of research are also provided. design and development of gamified e-learning systems. This
issue can be challenging when other researchers need to
Keywords—Gamification; e-learning systems; interaction
replicate their work. Furthermore, gamification is still in its
design; experimental evaluation
infancy and quickly developing requiring more focused and
I. INTRODUCTION thorough studies [14].
Learning is a complex issue that can be influenced by These issues need to be carefully addressed when applying
different factors including learner characteristics, learning gamification in learning. This paper bridges this gap by
content, learning environment and teaching style. Traditional offering a gamification design model that can be used to
learning approaches through classrooms can be rigid and develop different instances of gamified e-learning systems. The
unattractive to some learners given the time and place model is consisted of three major components including the
constraints. In contrast, e-learning systems provide the ability courseware module, the learner profile and the gamification
for learners to learn anytime, anywhere and offer different component. The paper also offers an evaluation of gamification
interactivity levels that may not well be supported by in an e-learning system, developed based on the proposed
traditional learning approaches. Moreover, learning content can model. A carefully designed and controlled experimental
be incorporated in e-learning systems with different types such evaluation was conducted with 44 participants in a real
as examples, simulations, problem-solving tasks and learning environment. The participants used the developed
explanations. These types can also be offered in different gamified e-learning system to learn some concepts related to a
multimedia formats including written and spoken material in course on computer security, the application domain. This
addition to videos and games. study primarily aims to explore the learning effectiveness of
the integration of gamification in e-learning systems. The main
Motivating learners to use such systems in order to enhance research question of this study is that does the integration of
their learning and gain some knowledge is a challenging task gamification in e-learning systems enhance learning?
[1]. Gamification is usually put forward as a proposed solution
to motivate, engage and support learners through their The rest of the paper is structured as follows. Section II
interaction with learning material [2]. The term „gamification‟ provides related work to gamification in e-learning systems.
was firstly introduced in the scientific community in 2010 [3]. Section III details the proposed gamification design model.
Since then, it is still widely adopted by researchers in different Section IV outlines the evaluation method. Section V offers the
research domains and areas including learning, business and results. Section VI concludes the paper highlighting the main
health [4]. Gamification can be defined as the use of game findings. Section VII points out to future directions of research.
elements, thinking and mechanisms in a non-game context [3].
II. RELATED WORK
E-learning systems can be designed to integrate different
Many attempts have been proposed to investigate
game elements such as points, badges, progress bars and levels
gamification in online-learning research. For instance, a
to support learners in having a gamified learning experience
blended study (classroom and online) conducted by [6] to
[5]. Gamification in learning should be differentiated form the
evaluate the learning effect of gamification yielding positive
204 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
205 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
206 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Measurement Tools
Learning gain was evaluated using three main tests: pre-
test, post-test and follow-up test. Two types of learning gain
are measured based on the tests. First, short-term learning gain
is measured as the difference between the post-test and the pre-
test in order to evaluate the immediate learning effectiveness of
gamification. Second, medium-term learning gain is calculated
as the difference between the follow-up test and the pre-test in
order to evaluate the sustained knowledge. Each test involves
25 items related to a course on computer security. Each item
has five multiple-choices with the fifth being „I don’t know!‟ in
order to reduce the chance of random speculating by Fig. 4. Summary for the Mean Scores of the Experimental Variables.
participants.
It should be noted that all tests are similar except for the Regarding the pre-test results, an independent sample t-test
ordering, arrangements and phrasing of both the items and the was conducted to compare the group who interacted with the
responses. A Cronbach alpha was also measured for the pre- gamified version of the e-learning system and the non-gamified
test, post-test and follow-up test having good reliability as 0.96, version. There was no statistically significant difference in the
0.82 and 0.83, respectively. scores for the gamified group (M=10.27, SD=3.92) and the
non-gamified group (M=10, SD=4.58); t(41)=-0.21, p=0.83.
C. Experimental Procedure These results suggest that all participants had the same
Participants were first welcomed and introduced to the knowledge level prior to the experiment which enabled for
process of the experiment after signing consent forms of more reasonable comparison between the two experimental
participation. Then, all participants were randomly assigned to conditions. Hence, any positive learning effect found was
one of two conditions: (A) a gamified version of the e-learning primarily caused by using the gamified e-learning system.
system (experimental) or (B) a non-gamified version (control). Reporting on the post-test results, there was a statistically
Participants in both A and B completed the pre-test, and then significant difference in the scores for the gamified group
studied some learning material related to computer security
(M=82.73, SD=9.83) and the non-gamified group (M=63,
using the system according to their assigned condition. The
SD=14.3); t(37)=-5.32, p=0.000, according to an independent
learning material in both conditions are similar and the
sample t-test. These results suggest that participants who used
difference is whether students were exposed to gamification or
the gamified e-learning system achieved better learning gain
not based on their assigned condition.
than those in the non-gamified version.
The learning task was completed in a computer laboratory
An independent sample t-test was also conducted
for three weeks where there were two experimental sessions
concerning the follow-up test. The results indicated a
per week. Each session was lasted about 120-180 minutes. At
statistically significant difference in the scores for the gamified
the end of all the experimental sessions, participants were
group (M=44.4, SD=10.4) and the non-gamified group
immediately directed to complete the post-test. Two to three
(M=29.8, SD=15.2); t(37)=-3.71, p=0.001. The learning gain
weeks after completing the main experiments, all participants scores according to the follow-up test in comparison to the
were invited to only complete the follow-up test. post-test scores have been decreased by about 50% in both
D. Data Analysis conditions because participants took the follow-up test after the
The data of the experiments were collected by the system experiment two to three weeks later affecting the sustained
automatically and stored in a database. Then, they were knowledge. Still, the results suggested positive findings related
converted into a format that enables for the data analysis to the experimental „gamified‟ condition.
process. Minitab 19 software was used for data analysis. The By obtaining the results related to the pre-test, post-test and
type of the tests performed is outlined when presenting the follow-up test, it is possible to test the hypotheses H1 and H2.
results in the following section. Regarding the short-term learning gain, calculated as the
difference between the post-test scores and the pre-test scores,
V. RESULTS an independent sample t-test was conducted. There was a
A number of 44 Male undergraduate students form the statistically significant difference in the short-term learning
College of Computer Science and Engineering, University of gain scores for the gamified group (M=72.5, SD=10.5) and the
Hail, Saudi Arabia completed the experiments. The gender non-gamified group (M=53, SD=14.7); t(37)=-5.04, p=0.000.
variable was controlled to eliminate any variance and According to the results, the hypothesis H1 can be confirmed.
confounding effect on the results. The conditions being It can be suggested that learning using the gamified version of
experimental „gamified‟ and control „non-gamified‟ were the e-learning system produces better (short-term) learning
balanced having 22 participants each. gain than learning using the non-gamified version.
There were five variables related to learning gain (pre-test, Reporting on the medium-term learning gain based on the
post-test, follow-up test, short-term learning gain and medium- conducted independent sample t-test, there was also a
term learning gain) in the experiment. Fig. 4 summarizes the statistically significant difference in the medium-term learning
mean score results for those variables. gain scores for the gamified group (M=34.1, SD=11.8) and the
207 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
non-gamified group (M=19.8, SD=15.9); t(37)=-3.39, p=0.002. improved to cater for collaborative and social learning besides
Based on the results, the hypothesis H2 can be confirmed. It is the integration of game elements in e-learning systems to
suggested that learning using the gamified version of the e- support active learning. An essential point to emphasize,
learning system produces better (medium-term) learning gain generally in online-learning research, is that gamified e-
than learning using the non-gamified version. learning systems should be designed following sound
instructional theories and models.
VI. CONCLUSION
REFERENCES
This paper contributed to the current literature by the [1] M. Sailer, J. U. Hense, S. K. Mayr, and H. Mandl, “How gamification
provision of the gamification design model. The model can be motivates: An experimental study of the effects of specific game design
used as a foundation to design different instances of gamified elements on psychological need satisfaction,” Computers in Human
e-learning systems. It is consisted of major components needed Behavior, vol. 69, pp. 371–380, 2017.
to develop such systems including the learner profile, the [2] I. V Osipov, E. Nikulchev, A. A. Volinsky, and A. Y. Prasikova, “Study
courseware module and the gamification component. An of gamification effectiveness in online e-learning systems,” International
Journal of advanced computer science and applications, vol. 6, no. 2, pp.
implementation of the model was also completed resulting into 71–77, 2015.
a gamified e-learning system in order to validate the proposed [3] S. Deterding, D. Dixon, R. Khaled, and L. Nacke, “From game design
model and to evaluate the learning effectiveness of elements to gamefulness: defining gamification,” in Proceedings of the
gamification. 15th international academic MindTrek conference: Envisioning future
media environments, 2011, pp. 9–15.
Moreover, a thorough experimental evaluation in terms of [4] G. Baptista and T. Oliveira, “Gamification and serious games: A
learning gain was conducted considering the limitations of literature meta-analysis and integrative model,” Computers in Human
published research. Learning gain was also reported not only Behavior, vol. 92, pp. 306–315, 2019.
according to the immediate (short-term) learning effect [5] I. Varannai, P. L. Sasvári, and A. Urbanovics, “The use of gamification
measured by a post-test taken by participants after completing in higher education: an empirical study,” International Journal of
the experiment, but also according to a delayed (medium-term) Advanced Computer Science and Applications, vol. 8, no. 10, pp. 1–6,
2017.
learning effect measured based on a follow-up test taken by
[6] M. Tan and K. F. Hew, “Incorporating meaningful gamification in a
participants two to three weeks later after completing the blended learning research methods class: Examining student learning,
experiment. The findings indicated that learning using the engagement, and affective outcomes,” Australasian Journal of
gamified version of the e-learning system produces better Educational Technology, vol. 32, no. 5, 2016.
(short-term and medium-term) learning gain than learning [7] D. Strmečki, A. Bernik, and D. Radošević, “Gamification in e-Learning:
using the non-gamified e-learning version. introducing gamified design elements into e-learning systems,” Journal
of Computer Science, vol. 11, no. 12, pp. 1108–1117, 2015.
VII. FUTURE WORK [8] J. Filippou, C. Cheong, and F. Cheong, “A Model to Investigate
Preference for Use of Gamification in a Learning Activity,” Australasian
Pointing out to future work, it is planned to conduct a Journal of Information Systems, vol. 22, 2018.
longer-term evaluation in a semester long duration adding [9] F. L. Khaleel, N. S. Ashaari, and T. S. M. T. Wook, “An empirical study
more learning resources and focusing on another application on gamification for learning programming language website,” Jurnal
domain in order to generalize the current findings with a larger Teknologi, vol. 81, no. 2, 2019.
sample size. It is also true that this paper reports on learning [10] T. Jagušt, I. Botički, and H.-J. So, “Examining competitive,
gain as an important variable; yet, other variables can be collaborative and adaptive gamification in young learners‟ math
considered in future experiments such as motivation and learning,” Computers & Education, vol. 125, pp. 444–457, 2018.
satisfaction since the study presented in this paper builds up the [11] K. Welbers, E. A. Konijn, C. Burgers, A. B. de Vaate, A. Eden, and B.
C. Brugman, “Gamification as a tool for engaging student learning: A
foundation to conduct more experiments by the proposed field experiment with a gamified app,” E-Learning and Digital Media,
model and by the careful experimental evaluation approach. vol. 16, no. 2, pp. 92–109, 2019.
Another potential future direction is to compare the effect [12] E. Kyewski and N. C. Krämer, “To gamify or not to gamify? An
experimental field study of the influence of badges on motivation,
of unique combination of different game elements on different activity, and performance in an online learning course,” Computers &
cognition and psychological factors with careful considerations Education, vol. 118, pp. 25–37, 2018.
to experimental design. Another possible research direction is [13] M.-B. Ibanez, A. Di-Serio, and C. Delgado-Kloos, “Gamification for
to investigate different personalities and learning style of engaging computer science students in learning activities: A case study,”
learners and to explore how they relate to certain game IEEE Transactions on learning technologies, vol. 7, no. 3, pp. 291–301,
elements. Also, the culture and the learning context can play an 2014.
important role in motivating or demotivating learners to use [14] J. Koivisto and J. Hamari, “The rise of motivational information
systems: A review of gamification research,” International Journal of
gamified e-learning systems. Therefore, more research is Information Management, vol. 45, pp. 191–210, 2019.
required to compare the effect of using gamified e-learning
[15] S. Subhash and E. A. Cudney, “Gamified learning in higher education:
systems in different learning contexts and cultures. A systematic review of the literature,” Computers in Human Behavior,
Additionally, researchers can conduct more research to explore vol. 87, pp. 192–206, 2018.
the extent to which learners are engaged in the learning process [16] J. A. Self, “The defining characteristics of intelligent tutoring systems
when using gamified systems in long-term evaluation studies. research: ITSs care, precisely,” International Journal of Artificial
Moreover, the proposed gamification model can further be Intelligence in Education, vol. 10, pp. 350–364, 1999.
208 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The widespread use of Internet has resulted in a challenges in sentiment analysis. First one is the opinion
massive number of websites, blogs and forums. People can easily words and the second one is the manner in which these
discuss with each other about different topics and products, and opinion words are expressed. The words used in an opinion
can leave reviews to help out others. This automatically leads to a can be positive (for a specific topic or issue) and negative at
necessity of having a system that may automatically extract the same time for some other issue. Whereas the manner
opinions from those comments or reviews to perform a strong depicts the style or tone in which the words are being used.
analysis. So, it may help out businesses to know the opinions of Either the words are used in a positive sense or the same
people about their products/services so they can make further words are used to taunt something. On the other side, language
improvements. Sentiment Analysis or Opinion Mining is the
is another hurdle for many of the smart and accurate sentiment
system that intelligently performs classification of sentiments by
extracting those opinions or sentiments from the given text (or
analysis systems. A number of systems have been developed
comments or reviews). This paper presents a thorough research to process English, but an intelligent multilingual system is an
work carried out on tweets’ sentiment analysis. An area-specific utmost requirement to cope with diversity of languages on
analysis is done to determine the polarity of extracted tweets for such social sites.
make an automatic classification that what recent news people Due to an emerging trend of communicating and sharing
have liked or disliked. The research is further extended to personal opinions, people participate in different events on
perform retweet analysis to describe the re-distribution of such sites particularly the ones owned by press media. This
reactions on a specific twitter post (or tweet).
participation comes in the form of comments and reviews on a
Keywords—Sentiment analysis; tweets; opinion mining particular news or report. These reviews and comments
embody a lot of subjective information. So, some intelligent
I. INTRODUCTION ways are required to be devised to extract meaningful
information from such opinions or comments. In this regard,
The study performed an area-specific sentiment analysis
the NLP systems propose a term Subjective Analysis to handle
on tweets to extract people‟s opinions or comments on the
the mentioned issue. This is an enclosed term that covers
recent news. The retweet analysis is also performed in order to
sentiments, emotions, opinions and evaluation. The two major
describe the re-distribution of reactions on a particular twitter
approaches that are incorporated for a meticulous and accurate
post.
analysis of viewer‟s responses are Sentiment Analysis and
The massive increase of social media (including Opinion Mining.
networking sites, blogs, forums, communities, etc.) on the web
The sentiment or opinion analysis for tweets or comments
has taken a new turn in the form of public opinion. The
is far more difficult and challenging than the systems that only
organizations, businesses and companies now consider public
collect user‟s feedback in the form of likes and dislikes. For
opinion (or feedback) an important aspect in decision making.
the tweets (or comments), it might be possible that people
But filtering out the necessary information from such social
have disliked something because of the unnecessary details
sites presents an issue that needs to be resolved. The
provided on a certain topic and they are not interested to get
credibility of social sites is very important to be analyzed as
into such details. This is the reason sentiment analysis is not
well as the diversity of languages on such social platforms is
merely to check each and every word, rather the system has to
another problem to deal with. This raises an utmost need of a
be guided to identify and extract what is beneficial for the
smart and intelligent analysis to filter out the desired public
analysis. It is a fact that tweets contain slang language, various
opinion on a certain issue.
internet writing styles, jokes, icons or commonly used web
Sentiment Analysis is the solution to the aforementioned phrases. All these things make syntax analysis complicated
problem which is an application of natural language and may lead to wrong classification of tweets. So, this is the
processing (NLP). This analysis system is capable of major requirement to devise a powerful sentiment analysis
extracting the opinions or views of public regarding a certain system that not only focuses on the extraction of information
topic. It consists of a system that collects opinions (or but also analyses the subjectivity of the tweet. Some of the
feedbacks) on a specific topic from various blogs, social major challenges for sentiment analysis systems are:
networking sites and reviewing panels. There are two major
209 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Named Entity Extraction–This is one of the major online communication is getting cheaper. Almost 70% of the
challenges. Being existed in every tweet, the named entity platforms are completely free and this leads to a massive
extraction refers to the extraction of main idea behind the participation of a layman in the form of reviews, comments
usage of any name, like the „Dreams of Mango People‟ is a and discussions regarding news updates, products, services,
term that is used to represent the thoughts and desires of etc. People from all over the world can give their opinions on
common people. such platforms, means the region‟s restriction is no more. This
attracts researchers and analysts who want to analyze such
Information Extraction–Another biggest challenge is the opinions and comments. This is where NLP plays its
extraction of meaningful information. „Was the tweet significant role in which such comments, opinions or tweets
informative or fake?‟ This is the question that a sentiment are analyzed to extract the useful information with the help of
analysis system is devised to answer accurately. computer programs. This entire process is referred to as
Sentiment Determination–It is about determining the sense „Opinion Mining‟.
in which the tweet has been posted. It is either in positive or The following paper focuses on the Sentiment Analysis
negative sense. Since sentiments are usually used in a subtle carried out on tweets. For this purpose, an area-specific
manner that makes it difficult to be analyzed from a single analysis is performed; the user‟s tweets are monitored on
word or sentence. For example: different Pakistani political as well as non-political issues. In
The private educational institutes are increasing fee by order to extract the user‟s feedback on a specific news, a
20%. But these institutes are providing a high-quality of mechanism (called Sentiment Analysis) is devised that
education. basically analyzes the polarity of those tweets (checking
positivity or negativity) to classify either people have liked it
The first part “The private educational institutes are or not.
increasing fee by 20%.” can be considered as a fact, while the
other part is based on a personal opinion. This makes it further A. Retweet Analysis Description
clear that identifying a single keyword for subjectivity is not First, confirm that you have the correct template for your
that easy as it seems. paper size. This template has been tailored for output on the
Parsing-In sentiment analysis, this phase is referred to as US-letter paper size. If you are using A4-sized paper, please
Semantic Parsing. This actually aims at highlighting the close this file and download the file “MSW_A4_format”.
semantic constituents (subject/object, verb/adjective) are It refers to the second major phase covered in this paper
identified. It is basically a formal analysis of a sentence and plays a significant role in classifying the tweets and their
performed by the computer. credibility. From different studies, it is found that important
Anaphora Resolution: This is the phase in which nouns news events or updates are retweeted (or shared) more.
and pronouns are identified. Basically, an anaphora relates an Similarly, the tweets with negative impact like news related to
expression to another one preceding it in the discourse. It natural disasters are retweeted more than others. This is why
presents a biggest challenge in Sentiment Analysis. Here the the retweet analysis description is performed to describe the
anaphoric expression can be explained by relating them with re-distribution of expressions or reactions on a particular tweet
the context of the whole sentence. Means it can be viewed as a or post.
summarization of the context by extracting different III. LITERATURE REVIEW
sentences. It helps out in Sentiment Analysis by identifying
which discourse identities are used repeatedly. The major concept behind Sentiment Analysis is to
determine the attitude of a user (or a writer or speaker)
Twitter is one of the widely used social networks that has a regarding a specific issue. Or it finds out the overall polarity
massive number of users including politicians, celebrities and of a document. Many of the work done in this realm is
companies. The given research focuses on the sentiment dependent upon machine learning approach. Due to the huge
analysis performed on Twitter data. The paper is organized as spread of user generated content through social media and
follows: forums, a massive work has been done on the sentiment
Section 2 provides a brief background about social media analysis for a social network. Also referred as Opinion
as well as Twitter and tweets. Section 3 gives a thorough Mining, this process is considered to be highly accurate way
literature review. Section 4 deals with the proposed technique to test credibility of any news or topic on social networks.
that is news credibility using Sentiment Analysis. Section 5 (Denecke, Kerstin, 2008) [1] devised a method that
gives implementation details while Section 6 provides automatically determines the polarity of sentences which are
performance measures, experimental results. And Section 7 not in the same language. This is called multilingual sentiment
gives conclusion with the recommendations for future work. analysis. First the text language is determined (if it was other
than English) and translated into English in order to carry out
II. BACKGROUND sentiment analysis in an easy way. The research also
With the advent of modern technologies and a gigantic determined the type of document, whether it is subjective or
increase of social network communication has made objective. Due to the diversity and complexity involved in
everything online. This is the reason With the advent of human languages, the research came across many difficulties.
modern technologies and a gigantic increase of social network Moreover, sentiment of text became more crucial, so the
communication has made everything online. This is the reason whole text was not analyzed.
210 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Most of the work done on Sentiment Analysis had utilized messages are full of slangs, short forms, abbreviations. This
the approaches of Support Vector Machine (SVM) and other makes it far difficult from other forms of sentiment analysis.
classifiers with binary uni-gram weights. (Paltoglou, Most of the Twitter sentiment analysis is done using machine
Georgios, and Mike Thelwall, 2010) [2] came up with the idea learning approach. The two major reasons for using Machine
that classification accuracy can be enhanced with information Learning techniques are:
retrieval scheme when more weighting schemes were used.
The work modeled a document as an unordered collection of 1) A huge amount of Twitter data is available for training
words called a bag of words. The approach came out to be datasets.
computationally efficient. More sophisticated term weighting 2) The test data is also available that is user-labeled for
functions were adopted from SMART retrieval system to sentiment with emoticons, so there is no need of manual
devise a probabilistic model. annotation of data for training.
(Caladoa, Edgar RochaaAlexandre et al., 2011) [3] had The emergence of Web 2.0 has brought improvements in
thoroughly discussed about the user contents posted on their the way people used to perceive Internet. Micro-blogging is
respective Twitter profiles as well as what type of people one of the most popular Web 2.0 applications that have
mostly post tweets. In order to carry out this research, they facilitated users to collaborate, share, discuss and leave their
first extracted a list of features. The list comprised of a feedbacks on different topics, news and products. Twitter
number of friends, followers, tweets and re-tweets. Moreover, being one of the most popular micro-blogging platforms, has
the user‟s tweeting behavior was tested incorporating feature been the hot area of research for many years. (Kontopoulos,
extraction method. Finally, the public behavior on real time Efstratios, et al., 2013) [6] had worked on the same discipline
news events was also examined. For this purpose, an and discussed about the ontology-based techniques for an
algorithm was devised that was used to extract re-tweet chains efficient sentiment analysis of Twitter posts. Their research is
and timestamp of messages. The algorithm performed user divided into two phases: a) creation of domain ontology and b)
profiling for which the limitation was that this profiling was sentiment analysis on a set of tweets based on the concepts
not enough for analyzing the behaviors of users posting tweets and properties of ontology. This work had utilized FCA
or replying to the tweets regarding news events. (Formal Concept Analysis) which is a mathematical data
analysis theory and typically used in knowledge representation
Another work performed on Sentiment Analysis by
and information management. They came across a difficulty in
(Takaoka, Kouichi, and Akiyo Nadamoto, 2011) [4] came up
advertising tweets in which an unpleasant ratio was involved.
with a new system titled „Words of Wisdom‟. They proposed
a system that was based on multi-dimensional sentiments (Montejo-Ráez, Arturo, et al., 2014) [7] performed
vector, in which two-system approach was utilized. First, a Sentiment Polarity Classification on Twitter posts using a
multi-dimensional vector (based on 10 categories of novel approach. They extracted a vector of weighted nodes
sentiments) was proposed, and then values for these proposed from the graph of WordNet. Then, these weights were used in
sentiments were calculated. A frequency vector was used to SentiWordNet in order to compute the final polarity. The
calculate the frequency of sentiments. The results came out to method proposed a non-supervised approach that was domain
be quite accurate and the authors proposed in future work that independent. Since Twitter publishes a vast range of
distances could also be calculated and adoption for news data information including political, economic, business and more
could also be incorporated. contexts, the scoring of posts is done as per the degree of
positive and negative opinions expressed therein. For this
Due to a rapid and huge spread of social networking and
purpose, SentiWordNet scores were combined with a random
blogging platforms, researchers are going deeper into this
walk analysis of the concepts found in the text over the
area. Online Opinion Mining is another form of sentiment
WordNet graph. Random Walk is an algorithm that was
analysis that has been greatly worked on but it is considered a
particularly used for mathematical formulization to perform
difficult form of sentiment analysis. (Haddi, Emma et al.,
random steps. A graph was also constructed for configuration
2013) [5] worked on the role of text pre-processing for
of the results that demonstrated different parts of WordNet
sentiment analysis and demonstrated how the sentiment
subgraph for the solid terms. The proposed method was
analysis can be further significantly improved by using
intended to calculate the global polarity of Twitter posts by
appropriate feature selections and representations. They made
expanding a fewer concepts that were in tweets. The limitation
use of Support Vector Machine (SVM) and performed a
of this work is that they didn‟t consider whole tweet text form
comparison of their accuracies with the accuracies acquired in
analysis. They also came up with the fact that taking too many
topic categorization. The research mainly focused on the
concepts will introduce noise in understanding the latent
product reviews (on social blogs) and determined the
semantic of the text.
importance of a product on the basis of those reviews. The
supervised learning for feature sections and representations (Abbasi, Mohammad-Ali, and Huan Liu., 2013) [8]
was used as the major technique and the accuracies achieved focused on the social media for information of the upcoming
were in the range of 75% to 83%. news events in the world. Since people are more interested in
getting first hand news, the paper worked on the same issue
Going beyond the researchers have shown a great interest
and proposed a method to measure user credibility in social
in Twitter-specific sentiment analysis. This kind of sentiment
media. For this purpose, they proposed CredRank algorithm
analysis is a bit different from the conventional one. Due to
that was solely devised to measure user credibility in social
the limited character length (up to 140 characters), Twitter
media. It analyzes the users‟ online behavior to measure the
211 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
said credibility. The proposed methodology worked in the online resources. After the collection, sentiment annotation
given steps: was performed in which the collected words were annotated
with corresponding sentiment scores. Then, a lexicon based
Detect and cluster coordinated (i.e. dependent) users sentiment analysis pipeline was created. This was comprised
together. of steps including: Tokenizer – that splitted text into very
Weight each cluster on the basis of the cluster‟s size. simple tokens. Sentence Splitter – that fragmented text into
sentences. POS Tagger – It produced a part of speech tag as an
Due to the anonymous and unmonitored mature of the annotation for each word or symbol. Gazetteer – This was the
Internet, the user generated content on Twitter might be basis of the proposed methodology, as each gazetteer entry
incredible. This incredibility leads researchers to work on when appeared in the text, it got marked and was assigned a
different ways to perform credibility analysis. (Gupta, Aditi, sentiment score accordingly. After passing the data through
and Ponnurangam Kumaraguru., 2012) [9] worked on this this pipeline, the JAPE rules (hand-coded grammar rules)
realm and performed credibility analysis on the information were devised. These rules were formed in two phases – phase
contain in a tweet corresponding to fourteen high impact news I formed word-level rules whereas phase II worked on
events of 2011 occurring globally. They conducted research sentence-level rules. Finally, the Groovy scripting processing
with the help of Regression Analysis through which they resource was utilized through which the number of positive
identified the important content and source-based features; and negative annotations (in a given text piece) were counted
helpful for predicating the credibility of information contained and an overall polarity was determined as well. The proposed
in a tweet. The ranking of tweets (according to their credibility method yielded around 60 – 70% accuracy rates for the initial
scores) was performed using the supervised machine learning version of lexicon-based sentiment analysis API. Although,
and feedback approach. The performance of the ranking the method did not come up that much efficient as most of the
algorithm was significantly improved when re-ranking ML based approaches. But it showed value and could be
strategy was applied. With all the data analyzed, it came out to combined with some ML based approach to produce a hybrid
be known that pn average 30% of total tweets posted about an system.
event comprised of situational information. While the
remaining 14% contained spam tweets. Whereas only 17% of (Sharma, Nitesh, et al, 2018) [11] designed a web-based
the total tweets with situational awareness information were application for performing sentiment analysis of live tweets.
credible. Pseudo Relevance Feedback (PRF) was used for re- Due to a massive use of social media, people use this platform
ranking purpose. This technique is also known as Blind for expressing their opinions on almost every topic. That is
Relevance Feedback and one of the most prominent re-ranking why the researchers came up with an idea of building a web-
techniques used in information retrieval tasks to improve the based application that not only performs sentiment analysis on
performance of re-ranking. PRF works by extracting K ranked live tweets but also visualizes the measured sentiments
documents and then re-rank them on the basis of denoted associated with keyword (hashtag, words or phrases) of
score. The algorithm extracted the most frequent unigrams Twitter messages. So, this enables users to measure the
from the top K tweets and re-rank them by utilizing the text sentiment of these messages in terms of geography. The
similarity between those most frequent unigrams and K application is designed on the framework „Flask‟ using Python
tweets. PRF was basically applied to the best set of results that programming language. This framework is built in a way that
was acquired by previous analysis (that is the ranking results user enters a keyword and application fetches the live tweets
using both message and source. Using the metric BM25, the (related to the entered keyword), extracts text from each tweet
text similarity between a tweet T and query set Q was and calculates the user location and sentiment for each tweet
determined for each event occurred. Around 50% of tweets on and finally plots the results on a map. Flask framework
an event are composed of the tweets which were related to the contains an initial configuration file that is used by Views
event but didn‟t provide any useful information. So, it was module. This module is responsible for rendering web pages
concluded that a ranking algorithm is based on both the user and communicating with API. It intercepts the incoming
properties and content. And it turns out to be very effective in requests and transfers the control to the back-end layer for
determining the credibility of information in these tweets. processing. As the result is ready, the views module generates
web pages. In order to extract tweets, Twitter streaming API is
(Amiri, Fatemeh et al., 2015) [10] had performed utilized that fetches live tweets corresponding to the entered
sentiment analysis for Persian text through lexicon-based keyword. Moreover, the attributes (like location, tweet text,
approach. Since a very little amount of work has been done so followers count, friends count, tweet time) are extracted using
far on Persian language, therefore in order to gain insights the meta data. For extracting the user location, a parser is
from different online sites and social media, sentiment created that maps user countries to the countries extracted
analysis was performed. But the researchers came up with a from data. In this regard, there is one additional feature
novel approach of incorporating a manually created lexicon particularly for the US residents that the app generates their
that was enriched with sentiment scores, coupled with hand- state wise plotting on map as well. For sentiment calculation,
coded grammar rules. The work also addressed some of the Python‟s text-blob library is incorporated. The sentiment score
Persian language issues including difference between formal assigned to words range from -1 to +1 and a polarity score of
and informal writing styles, context sensitivity, complexity 0 is termed as neutral sentiment. An in-depth analysis of data
due to frequent morphological operations, lexicon intricacy is also provided in a way that system also calculates the mean
etc. To perform the proposed idea, they first manually and weighted polarity values. Furthermore, the system also
collected Persian adjectives, words and expressions from two calculates the number of tweets from each country and from
212 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
each within state of the USA. Map plotting is done with plotly Word2Vec and Glo Ve. Furthermore, Word2Vec includes four
and Tableau. This application is unique in a sense that it can models – two from Skip-Gram and two from Continuous Bag
perform sentiment analysis on live tweets as well as on of Words (CBOW). The parameters used across all of the
previously gathered tweets stored in a database. When the models were: minimum word count (means the number of
system was evaluated using a search string #Watercrisis, the times a word is present in the corpus), word vector dimension
results were found that a total of 1,164 tweets were extracted size and window context size. Moreover, Word2Vec and Glo
related to this hashtag (or keyword) from around the world. Ve required different hyperparamters including Learning rate,
And their respective sentiment scores were successfully Weighting, Iteration and more. Whereas, ELM required only
calculated and visualized on map as well. one hyperparamter that is Regularization. For generating ELM
word embeddings, the foremost step is to build a vocabulary
(Lauren, Paula, et al., 2018) [12] have conducted a and its major parameter is minimum word count that finds out
massive research by generating word embeddings from an how many times a word should appear in the corpus to be
extreme learning machine for the sentiment analysis and included in the vocabulary. Then, frequency counts are
sequence labeling tasks. Word embeddings are basically the rendered for each of the word pairs from the training corpus.
low-dimensional distributed word representations for a set of The window context size is utilized that determines how far to
language modeling and feature learning techniques. The words the left and right each word has to be used for the co-
or phrases from the designed vocabulary are mapped on to occurrence counts. A weighting scheme is incorporated to get
vectors of real numbers (in low dimensional space). This a numerical representation of context, so the closer words are
research focuses on ELM based word embeddings for the given higher weight and the distant ones are assigned with
sentiment analysis and sequence labeling tasks. There are lower weight values. The word context matrix is a square
already different models for generating word embeddings; matrix with the dimensions matching to the size of
Word2Vec and Global Vectors (Glo Ve) are the popular of vocabulary. The square root transformation of the word-
them. In this research, they have also done a comparative context matrix with l2- normalization is done before applying
study in which ELM based word embeddings are compared
ELM. The word vector dimension size corresponds to the
with the aforementioned models. Both models use word-pair number of neurons in the hidden layer of ELM. For this
co-occurrences (also called as Word-Context Matrix) for purpose, the MATLAB ELM Autoencoder source files were
learning the low-dimensional representation of words. used. The study has performed both intrinsic and extrinsic
Word2Vec computes this matrix one row at a time while Glo evaluation The intrinsic evaluation means word embeddings
Ve computes the matrix at once and then applies matrix are evaluated on the basis of semantic relatedness whereas in
factorization. The first model is known as Predict-based as it extrinsic evaluation the word embeddings are assessed in the
performs line-by-line computation, whereas the other one is downstream NL Task like text categorization. The results on
called as Count-based method because the word pair counts sentiment analysis task are evaluated using Precision, F1 and
are performed all at once in the first. Their results are Recall measures. For the sequence labeling task, Precision and
comparable, but Word2Vec takes more time as it needs to Recall are measured and F1- score is applied utilizing the
train the neural network also, while Glo Ve consumes more Precision and Reall equations. In sentiment analysis task, the
memory. On the other hand, Extreme Learning Machine results of F1-score demonstrate that ELM word embeddings
(ELM) is actually a type of feed forward neural network and are competitive with Skip-Gram and Glo Ve word
its efficiency is attributed to the random non-updated hidden embeddings. And for the sequence labeling, Precision, Recall
layer weights as well as the efficient learning of output layer and F1-score are averaged across 20 executions using the test
weights. This study has utilized an Autoencoder architecture set incorporating the RNN for six models. The experimental
based on ELM for its feature learning functionality. results show that CBOWHS model did well overall in
Autoencoder itself is a type of neural network that performs comparison of other three Word2Vec models. While the ELM
feature learning by compressing the input feature vector (in word embeddings show a slightly better average in terms of
the hidden layer) and then decompressing it back in the output F1-score as compared to other five models.
layer. This compression feature is very useful because in
generating word embeddings using an ELM, the compressed (Shirsat, Vishal S et al., 2019) [13] discussed about
representation is desired because word embeddings sentence level sentiment identification by performing research
correspond to the low-dimensional representation of Word- on news articles. The data (of news articles) was extracted
Context matrix. Furthermore, a recurrent neural network from BBC news and sentence-level negation identification
(RNN) is also utilized in this study for the sequence labeling was basically applied. Sentiment Analysis is basically
task. An RNN is a feed-forward neural network that contains categorized as: Document level sentiment analysis in which
recurrent connections back to the previous layer. This feature the polarity of entire document is determined, Sentence-level
is useful for processing sequential data. They have sentiment analysis in which each sentence is analyzed and
incorporated Elman RNN for this study. Then for sentiment polarity is determined. And Aspect-level sentiment analysis
analysis task (to assess the models of word embeddings), that applies analysis on objects and their respective features.
Logistic Regression is applied as a classification algorithm in This study uses sentence-level sentiment analysis on news
which the outcome Y=1 means positive result, whereas Y=0 articles using Machine Learning Algorithms Support Vector
means the outcome is negative. For conducting research, three Machine and Naïve Bayes. The proposed methodology
separate datasets are utilized in which two datasets belong to consists of five major steps in which the foremost step was to
Sentiment Analysis task and one for Sequence Labeling task. preprocess the dataset in which the irrelevant text (in news
Word embeddings were generated using three models: ELM, articles) like HTML tags, advertisements and scripts are
213 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
removed. So, the data is prepared for text classification in A. Data Collection
further steps. The practical implementation involves applying Data extraction and collection are the major parts of this
Machine Learning algorithms to perform the classification research. Extraction is actually a time-consuming process. In
task. Naïve Bayes algorithm is one of them that determines the order to perform this task, first an account was created on
probability of an occurrence given the probability of another Twitter and an application was also created for authentication
occurrence that has already occurred. Another algorithm that to extract the required tweets. Twitter API was incorporated to
is used in study is Support Vector Machine which is non- achieve this work. API is an application programming
probabilistic algorithm and works well for both sequential and interface, in which a set of protocols and tools for
non-sequential data. After preprocessing of data, the next step development of different software applications is defined. This
is to apply Stemming in which the entire document is API is open source from where the code (related to Twitter
transformed into lower case to gain uniformity. Stemming is data) is extracted and can further be used in different respects
basically the truncation of a word to its root form. After this, of research work.
the Term Document Matrix is determined, this matrix defines
the frequency of terms that appear in the preprocessed dataset. B. Proposed Architecture
The rows of matrix are collection and columns correspond to The proposed architecture of the system is divided into
the related terms. Then, sentiment score generation is two steps: Sentiment Analysis and Retweet Analysis. Let‟s
performed with the aid of positive and negative dictionary. For analyze both steps here:
this purpose, each word in the preprocessed dataset is
compared to the word in dictionary to determine whether it is Sentiment Analysis Architecture–The proposed
positive or negative. Finally, the Naïve Bayes and Support architecture of sentiment analysis depicts the entire scenario of
Vector Machine algorithms are applied to perform sentiment analysis that how it works on the provided data. The
classification and estimate the accuracy. The experiment was major performance measures applied for the manual
conducted on the five categories of news articles including annotation (during sentiment analysis) are:
Entertainment, Business, Politics, Sports and Technology. Confusion Matrix–Also referred to as contingency table,
They utilized Bing Liu dictionary for determining positive and matching matrix or errors matrix; this measure is applied on
negative words. This dictionary contains 2006 positive and both supervised and unsupervised learning systems. This
4783 negative words. The experimental results demonstrated measure is used in order to have a visual overview for the
that Naïve Bayes achieved an accuracy of 96.46% for overall performance of the model used in work. Actually, the
Entertainment category and the lowest accuracy was 92.63% matrix compares the actual class values with the predicted
that was for Business category. While Support Vector ones. This way, it analyzes whether the presented model is
Machine achieved highest accuracy 94.16% for Politics confused between the two classes or not. The comparison of
category and the lowest as 69.01% in Sports category. predicted values with the ground true values gives a clear
(Iqbal, Farkhund, et al., 2019) [14] came up with a novel picture about the proposed model. The matrix is drawn in the
approach in which they designed and developed a hybrid form of table shown in Fig. 1.
framework for performing sentiment analysis. This framework It is clear from the matrix that there are two classes
combined ML based algorithm along with lexical database to positive and negative. But they are further divided in boxes
automatically analyze the online content (including reviews, and termed as TP, FP, TN and FN
social media). Then a Genetic algorithm based feature
reduction solution is provided through which further accuracy TP corresponds to the true positive terms. These are the
and scalability is achieved. Moreover, they have proposed a values that are positive in actual and the proposed model has
cross-disciplinary area of geopolitics to which they have also predicted them as positive.
applied the proposed framework as a case study application. It FN means false negative. Means these terms are actually
is a complete unique approach in which they tested and negative but the proposed model did not predict them as
experimented the accuracy of the proposed approach by negative.
applying it to the topics like terrorism, global conflicts, etc.
TN represents the values that are negative in actual and the
IV. METHODOLOGY proposed model has also predicted them to be negative.
This section gives a detailed discussion about the proposed FP represents the values that are positive in actual but the
methods for performing the desired task. The main focus of proposed model didn‟t predict them to be positive.
the work is to perform sentiment analysis on the public
opinions and reviews on Twitter on the daily news updates. Precision–Precision is basically defined as the ratio of all
Therefore, the most important part is to determine how people those instances which are correctly predicted by the classifier
react about a certain news or topic. In this section, the as positive. Although, it is a performance measure for binary
proposed methodology gives an overview about how data is classification (and we have used multi classification), but it
extracted for analysis as well as the method applied to perform can be used for multi class classification. The formula is
extraction. The next phase deals with data pre-processing and defined below:
how the application shows accuracy in results. In this regard, a
tool is proposed that is used to extract data from Twitter and
analyzes it according to the applied algorithm.
214 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
215 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
216 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Trending Topic: It corresponds to the topics or trends that before applying performance measures and it is manually done
are currently in search. These trending topics are uploaded by checking twitter posts and the sentiments against each
daily on Twitter to give the users an overview about the trending topic and then applying performance measures.
trending news or updates.
Manual Annotation on Training Data: Twitter posts are
Keyword Matching: In order to extract tweets, a specific manually labeled and then results achieved are compared with
keyword is given to the application. Then, application starts the results provided by the application. The performance
searching tweets (on the website) related to that entered measures are used to measure accuracy and all comparisons
keyword. As the relevant tweets are found, the application are made on the basis of these measures. A few examples of
connects with Twitter API to download those tweets. Then, positive, negative and neutral tweets which are annotated are
API performs authentication of user with oath keys (that are provided in Tables I and II.
special keys provided to Twitter users who want to download
data from the website). Further analysis is performed after the Classification Results: SVM classifier was utilized to get
downloading of data. the desired results. The obtained accuracy is measured to be
95%.
Sentiment Analysis: Basically, there are three types of
Sentiment Analysis: i) Word based Sentiment Analysis, TABLE. I. EXAMPLES OF POSITIVE POSTS
ii) Sentence based Sentiment Analysis, and iii) Document
based Sentiment Analysis. Our study focuses on sentiment Twitter Posts Category
analysis of sentence as well as words. We have researched #PTI will make #KARACHI the city of lights again in Naya
about the reactions, emotions and feelings of people on #Pakistan InshAllah #PTI4Karachi Positive
different news (related to Pakistan). After matching and #bekhaufJunooniKarachiwalay
extracting words related to the feelings, sentiment analysis
categorizes the extracted words into three categories: Positive, I congratulate @ImranKhanPTI ; other leaders of PTI on
Positive
Negative and Neutral. successful public meeting in Karachi this evening.
Positive Sentiment: The reactions (or comments) on the Happy Defence Day...Long Live Pakistan @MHaris_
news (or Twitter post) showing happiness and positivity are @__Shaikho @jdk_update @OyeeParkashaayy @KyunBatow Positive
categorized as positive sentiments. We have, at the backend, @jiyaleem
incorporated separate dictionaries with both positive and
negative words. Moreover, two languages are considered, Proud Soil, Proud Nation with Determined Force…… National
positive
English and Urdu. Since, we have performed research on Air Defence Day…….Pakistan Zindabad
Pakistani news based Twitter posts, so most of the tweets are
in Urdu language. The most common positive words are Allah protect our families in Multan and the surrounding area
positive
from the flood threat@MultanInFlood
happy, nice, good, well-done, thanks, welcome, I like it and
more. PML-N & CM Punjab are famous for progressive work in
positive
Negative Sentiment: The same procedure (as used in Lahore. One big rain disclosed all efficiency in Lahore.
positive sentiments) is applied for negative sentiments. If an
extracted post has more negative words than positive and TABLE. II. EXAMPLES OF NEGATIVE POSTS
neutral, the overall post is considered negative. The most
common negative words are worried, sad, sorrow, angry, Twitter Posts Category
crying, bad etc.
Ahmad_Noorani: Geo ISD Real fascist face of PTI.
Neutral Sentiment: The sentiments that don‟t show any MQM much better than PTI. Karachi ppl wil reject PTI negative
particular feelings like happy, sad, angry, etc. are the neutral terrorists in next elections..
sentiments. It gets difficult to identify the neutral tweets
because they contain both positive and negative words. If post Failed show by PTI in Karachi yesterday, hardly 20K
negative
contains maximum number of positive words, the polarity is people came. #BurnolForPTI
measured positive and if maximum number of negative words,
After So Much Rain Still 14 Hour LoadShedding in
the polarity is negative. Otherwise the post is considered to Lahore camp; in result of that People have attacked Lesco negative
have neutral polarity. office Today burned its furniture !!
Polarity Check: To measure polarity of the tweet text, the Heavy rain starts again in Lahore with a flood warning
words are compared with both dictionaries (English and Urdu) issued. - Bubblews http://t.co/VL8at9ytTT via negative
in parallel. As the word is found in a dictionary, then polarity @GoBubblews
is classified, in the second step, as positive, negative or
neutral. Then number of words are counted and on the basis of RT @ABKool: PMLN brought gullu butts in police to
this count, polarity of overall post (i.e. positive, negative or attack azadi march and CM KP called people from KP to negative
come protect Khan
neutral) is assigned.
Training Data: In order to achieve accuracy in results, it is Atleast five armed militants attacked Astana Fazal in
Sargodha on September 6, Pakistan Defence Day. - negative
required to train application on specific performance http://t.co/11Beh1ejLQ
measures. In this regard, human annotation is performed
217 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
VI. EXPERIMENTAL RESULTS AND DISCUSSION Sentiment analysis: Sentiment analysis on extracted tweets
This section discusses the experimental results with is then performed that shows positive, negative or neutral
complete details including tables and graphs. The step by step tweets as per the given polarities. For this purpose, this
process is given below: application analyzes the polarity of words in each tweet. And
the number of positive and negative words are counted. With
Keywords selection: This is the step in which user goes to higher number of positive words, the polarity is defined to be
Twitter‟s homepage and searches for the trending topics. As positive and similarly with negative words. The figure and
per the location (of the user) specified, Twitter generates the table shown below specify this step clearly.
trending topics. The information is presented in Table III.
Retweet analysis: Retweet Analysis is performed in
Extraction of tweets: The keywords are added through parallel to show the credibility of results. This analysis works
analysis box. Press the create button to add analysis word in on the concept that if a specific keyword has 35 or more
the analysis drop down menu. As the Run Analysis button is retweets in a particular time interval, that topic is considered
clicked, the program extracts tweets as per the entered to be an interesting one that‟s why more people are interesting
keyword. The tweets are generated in the form of bundles like in that update. So, it is labeled as a credible news update. This
300 tweets in one bundle. is presented in Table IV and can be visualized in Fig. 3.
218 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Dataset details
Attributes no. Five classes
Characteristics of attribute Integers
Missing values Nil
Predicted variables Positives and negative tweets
219 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Topic= “Rain in Lahore”-The polarity found for this news After collecting the tweets related to news updates, they were
update was negative. In the month of September, heavy rain analyzed to determine their type as per their respective
occurred that was quite destructive for Lahore and the cities polarities. The chances of noisy data were reduced by
around Punjab. The application gave 298 tweets in which 257 downloading the tweets through keyword matching process.
were negative tweets and merely 9 were positive ones. The
accuracy results (using performance measures) are provided in One point is to be cleared that the tweets extracted are all
Table VIII. Statistics related to confusion matrix are presented related to the trending news events means the topics that are in
in Fig. 6. great discussion and appear on Twitter‟s homepage on a daily
basis. The reviews or reactions on such tweets are used to
Overall Accuracy: 81.57% determine the polarity as well as credibility of the news
Kappa coefficient: 0.513 updates. It was found during the study that sentiment analysis
Sensitivity: 54.55% of tweets can be performed independently without concern of
Specificity: 92.59% their context. But feature extraction is the crucial element that
True Positive = 30, True Negative = 125, False Positive = 10, matters a lot. In this regard, uni-gram and bi-gram proved to
False Negative = 25 be the better features than other ones for performing reliable
sentiment analysis.
The overall accuracy achieved for each of the extracted
topics depicts the authenticity of the proposed system and the In the next phase, the re-tweet analysis was another
potential it has to get enhanced for more advanced results in challenging task to be performed. By setting a threshold value
future. >= 35, the approach worked quite better. This analysis was
performed in parallel and credibility of the news events was
TABLE. VIII. PRECISION AND RECALL VALUES (RAIN IN LAHORE)
proved on the basis of re-tweet analysis. If a particular
keyword has 35 or more retweets, the topic was considered to
Classification Procedure be interesting. And more people were found to be taking
Class 1 Class 2
overall Accuracy(precision) interest in that topic. This further proved that news is credible.
Class 1 30 10 40 75% There were some challenges faced during this study. The
Class 2 25 125 150 83.33% collection of tweets related to news updates was quite a
challenging and difficult task. After overcoming this
Truth overall 55 135 190
challenge, we came across with another challenge that was in
User accuracy the form of manual annotation. In manual annotation, it is
54.54% 92.59%
(recall) difficult to decide whether a given tweet is positive, negative
or neutral. Since the tweets are related to trending news events
and there is always a possibility that tweet is dependent on the
context that makes it further difficult. In this regard, a
subjective approach is required because annotation is highly
subjective. The limit of 140 characters in a tweet also brings
some difficulties, as people find difficulty in expressing their
views or opinions in such a restricted limit. So, they use
abbreviations, short forms and slangs that make sentiment
analysis further challenging.
More future work and studies can be done in this realm.
By proposing more complex algorithms and utilizing
appropriate measures, a more scalable approach can be
proposed. The advanced techniques can be used to enhance
the work for resolving issues like detecting fraudulent or
criminal activities on social media. Moreover, we focused on
English and Urdu languages while ignoring the slang terms.
The future work may incorporate the technique that may be
capable of analyzing the slang terms as well.
Fig. 6. Confusion Matrix Values (Rain in Lahore).
REFERENCES
VII. CONCLUSION [1] Denecke, Kerstin, “Using sentiwordnet for multilingual sentiment
analysis.” IEEE 24th International Conference on Data Engineering
The research is aimed at solving a practical problem of Workshop, 2008.
Sentiment Analysis of Twitter posts related to news updates. [2] Paltoglou, Georgios, and Mike Thelwall, “A study of information
We have presented the background, related literature review retrieval weighting schemes for sentiment analysis” Proceedings of the
as well as possible approaches, techniques, features and 48th annual meeting of the association for computational linguistics.
Association for Computational Linguistics, 2010.
assumptions for Sentiment Analysis about news events. The
[3] Caladoa, Edgar RochaaAlexandre P. Franciscoa Pável, and H. Sofia-
data collection process is also thoroughly specified that we Pintoa.” User profiling on Twitter”, 2011.
have collected the required data with the help of Twitter APIs.
220 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[4] Takaoka, Kouichi, and Akiyo NadamotoK. Elissa, “Title of paper if [10] Amiri, Fatemeh, Simon Scerri, and Mohammadhassan Khodashahi.”
known,” unpublished.” Words-of-wisdom search based on multi- Lexicon-based sentiment analysis for Persian Text.”. Proceedings of the
dimensional sentiment vector”, Proceedings of the 13th International International Conference Recent Advances in Natural Language
Conference on Information Integration and Web-based Applications and Processing. 2015.
Services. ACM, 2011. [11] Sharma, Nitesh and Pabreja, Rachit and Yaqub, Ussama and Atluri,
[5] Haddi, Emma, Xiaohui Liu, and Yong Shi.” The role of text pre- Vijayalakshmi and Chun, Soon and Vaidya, Jaideep.” Web-based
processing in sentiment analysis.”, Procedia Computer Science, 2013. application for sentiment analysis of live tweets.”. Proceedings of the
[6] Kontopoulos, Efstratios and Berberidis, Christos and Dergiades, 19th Annual International Conference on Digital Government Research:
Theologos and Bassiliades, Nick.” Ontology-based sentiment analysis of Governance in the Data Age..2018.
twitter posts.”, Expert systems with applications, 2013. [12] Lauren, Paula and Qu, Guangzhi and Yang, Jucheng and Watta, Paul
[7] Montejo-Raez, Arturo and Martinez-C‟amara, Eugenio and Martin- and Huang, Guang-Bin and Lendasse, Amaury.” Generating word
Valdivia, M Teresa and Urena-Lopez, L Alfonso.” Ranked wordnet embeddings from an extreme learning machine for sentiment analysis
graph for sentiment polarity classification in twitter”, Computer Speech and sequence labeling tasks.”. Cognitive Computation.2018.
& Languag, 2014. [13] Shirsat, Vishal S., Rajkumar S. Jagdale, and Sachin N. Deshmukh.
[8] Abbasi, Mohammad-Ali, and Huan Liu.” Measuring user credibility in "Sentence Level Sentiment Identification and Calculation from News
social media.”, International Conference on Social Computing, Articles Using Machine Learning Techniques." Computing,
Behavioral-Cultural Modeling, and Prediction., 2013. Communication and Signal Processing. Springer, Singapore, 2019. 371-
376.
[9] Gupta, Aditi, and Ponnurangam Kumaraguru.” Credibility ranking of
tweets during high impact events.”, Proceedings of the 1st workshop on [14] Iqbal, Farkhund and Hashmi, Jahanzeb Maqbool and Fung, Benjamin
privacy and security in online social media. ACM. 2012. CM and Batool, Rabia and Khattak, Asad Masood and Aleem, Saiqa and
Hung, Patrick CK.” A Hybrid Framework for Sentiment Analysis using
Genetic Algorithm based Feature Reduction.”.IEEE Access.2019.
221 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—One of the major culprits that faces Mobile Ad-hoc Since it is clear that broadcasting plays a very critical role
networks (MANET) is broadcasting, which constitutes a very in network discovery and assists the nodes in MANETs in
important part of the infrastructure of such networks. This discovering their neighborhood [6], optimizing it constitutes a
paper presents a nested genetic algorithm (GA) technique with major step as it will save both energy and time, especially since
fuzzy logic-based fitness that optimizes the broadcasting most of the devices in the network have limited energy as they
capability of such networks. While normally the optimization of are battery powered.
broadcasting is considered as a multi-objective problem with
various output parameters that require tuning, the proposed Due to the previously mentioned limitations, a key threat
system taps another approach that focuses on a single output known as node ‘selfish behavior’ arises in the network, in
parameter, which is the network reachability time. This is the which the nodes purposely tend to drop the messages that do
time required for the data to reach a certain percentage of not target it, in an effort to save its energy [7] [8]. In other
connected clients in the network. The time is optimized by tuning words, the nodes are not encouraged to contribute to the
different decision parameters of the Delayed Flooding with forwarding process. This kind of self-regarding behavior
Cumulative Neighborhood (DFCN) broadcasting protocol. The negatively impacts the network because, as already stated,
proposed system is developed and simulated with the help of the there is no solid infrastructure in MANET and all the nodes
Madhoc network simulator and is applied on different realistic rely on the cooperation of other nodes in the network to deliver
real-life scenarios. The results reveal that the reachability time and forward their messages. Delayed Flooding with
responds well to the suggested system and shows that each
Cumulative Neighbors (DFCN) is a broadcasting protocol that
scenario responds differently to the tuning of decision
parameters.
can handle this behavior and, at the same time, can reduce the
number of packets that need forwarding with minimal punitive
Keywords—Broadcasting; DFCN; fuzzy logic; genetic actions on the final coverage [9]. This is achieved by dropping
algorithms; Madhoc simulator; MANET the forwarded message when enough of the neighborhood
devices have already got it. Also, once a node decides to
I. INTRODUCTION forward a certain packet, it waits for a specified amount of time
Mobile Ad-hoc Networks (MANETs) are dynamic types of before executing this action, which is then canceled if another
network consisting of an uncontrolled setup of end-point node in the network actually forwards the message [10].
communication devices known as terminals, which are able of The work proposed in this paper tackles a specific type of
arbitrarily connecting with each other without the need of a MANET, known as Metropolitan Mobile Ad-Hoc Networks,
base station or a fixed infrastructure [1]. The types of devices which is characterized by a disparate density that is
that are usually found in MANETs are laptops and continuously changing, whereas highly dense areas can swing
smartphones equipped with limited range wireless technologies from being active to inactive over short periods of time.
such as Bluetooth and WiFi (802.11). This, in turn, limits the Because creating a real testbed for this type of network is very
communication capability of such devices, but allows them to costly and challenging, and might also lack the reproducibility
move while communicating. factor, it was decided that the best approach to handle it is by
This makes the MANET very unpredictable as it needs to means of a simulation framework. The Madhoc [11] simulator
continuously self-reconfigure itself to accommodate these has been selected to achieve this. An evolutionary algorithm-
dynamic changes [2]. This is considered a major drawback for based technique that combines nested GA with fuzzy-based
the efficiency and effectiveness of the MANETs and, by failing fitness is proposed and implemented. The technique integrates
to readjust, link breakage will start to take place and some of the Madhoc simulator in its core and considers DFCN
the routes can become undiscoverable [3]. For the devices to be optimization over multiple real life mobility scenarios.
able to reach a certain destination, they start sending route The rest of this paper is organized as follows. Section II
discovery requests to their neighboring nodes [4] which, in introduces the Madhoc simulator and gives an insight about its
turn, do the same thing. This results in the network being capabilities and the different modes of operations. In
overwhelmed with an extreme amount of broadcast traffic Section III, a review of the related work concerning the
known as a broadcasting storm [5]. optimization of broadcasting techniques in MANETs is
presented. Section IV highlights the main problems that this
research aims to solve. Section V demonstrates the algorithms
222 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
and techniques used to solve the problem. Section VI shows Based Detection (RTBD) technique was simulated that can
the obtained results and discusses them. Finally, Section VII efficiently detect selfish nodes in MANET. The main focus of
concludes this work and proposes the potential future work. this work was to accelerate the detection of misbehaving
selfish nodes. The proposed method managed to diminish the
II. MADHOC SIMULATOR overhead, latency and overhead ratio which improved the
Madhoc is a metropolitan MANET simulator completely broadcasting performance of the MANET. However, the
written in Java and available to use publicly [12] on the authors did not demonstrate how the acquired security could be
author’s website [13]. The simulator provides the ability to transferred to the neighboring nodes in the network so that they
simulate MANET using different parameters and real-life could avoid being compromised by the selfish nodes detected
constraints such as working area size, mobility speed, wall by RTBD, meaning that the technique is not scalable on larger
thickness, etc. It also supports many different wireless networks and the performance will be degraded. Another key
technologies (e.g. WiFi, Bluetooth, GSM, etc.). Most focus in the literature is intelligent rebroadcasting techniques
importantly, it implements the full DFCN broadcasting that reduce the overhead by estimating the usefulness of
protocol with all the required decision parameters to optimize rebroadcasts and the probability of causing a collision. S. S.
it. Madhoc can be executed as a standalone application or as an Basurra et al. [17] discussed a Zone based Routing with
Application Programming Interface (API). Parallel Collision Guided Broadcasting Protocol (ZCG) to
reduce redundant broadcasting and to accelerate the path
To be able to collect the required statistics and results, a discovery process. The authors compared ZCG with two other
Madhoc monitor class is used. A monitor is not a part of the techniques, Dynamic Source Routing (DSR) and Adhoc
physical network and does not have an instance in real Ondemand Distance Vector Routing (AODV). It was
networks, and is regarded to as an abstraction entity that only concluded that ZCG can speed up the routing process in
exists at simulation level. It mainly aims at maintaining a MANET due to its on-demand parallel collision guided
global perspective on all nodes and for carrying out the broadcasting. However, the proposed method lacked
required operations such as node deployment and initialization. distribution fairness among the nodes and did not protect zone
It mainly serves as an observer of the Ad-hoc decentralized members from selfish behavior attributed to the Zone Leader.
process. Another major attribute of the Madhoc simulator is Another interesting finding in the literature is the clustering of
that it does not use an event-driven simulation architecture, but MANETs as a mean to reduce the complexity of the routing
instead, the simulator’s kernel iterates upon a discrete time table. M. Ahmad et al. [18] provided a comprehensive survey
domain, where the distance between two intervals is known as about the different clustering algorithms that address this issue.
the resolution. It concluded that the effectiveness of the clustering algorithms
This parameter is defined by the user and should be fixed depends on a set of specific parameters, which the nodes are
throughout all the related applications to guarantee comparable remaining power, the relative mobility, the overhead data, the
and consistent results. The higher this value is, the less accurate trust value, and the node reputation.
the simulation will become. This value should be carefully
IV. PROBLEM STATEMENT
used according to the required application. In the case of
DFCN, this value must be at least twice lower than the In order to optimize the DFCN protocol, multiple decision
maximum RAD, otherwise the benefits of using RAD will be parameters need to be considered. These parameters dictate
completely lost. how DFCN operates and they characterize the search space.
Since the optimization heavily relies on each specific scenario,
Another important factor to consider while choosing the an individual optimization trend is expected for each scenario.
resolution is the mobility scheme of the nodes, the resolution
must be small enough to make sure that the nodes move in The reachability time tr is the output benchmark that is used
reasonable steps, otherwise, some connections that could have to measure the optimization result. It is the amount of time
taken place in real life would not be simulated. required for the network to reach a certain number of pre-
defined nodes. The goal of this research is to optimize the
III. RELATED WORK DFCN parameters to decrease the reachability time of the
In the literature, most research has been dedicated to nodes inside the MANET. The problem is formulated as
solving the broadcasting issues by using a multitude of follows:
different methods. Evolutionary multi-objective approaches 𝑚: 𝑖𝑛𝑠𝑡𝑎𝑛𝑐𝑒 𝑜𝑓 𝑀𝑎𝑑ℎ𝑜𝑐 𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑜𝑟, 𝑡𝑟 : 𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑡𝑖𝑚𝑒.
have been proven to be effective in solving broadcasting
problems [14], however, they suffer from time and 𝑡𝑟 = 𝑚(𝐿𝑜𝑤𝑒𝑟𝑅𝐴𝐷,
performance issues [15]. Other methods focus on combinatory 𝑈𝑝𝑝𝑒𝑟𝑅𝐴𝐷, 𝑃𝑟𝑜𝐷, 𝑀𝑖𝑛𝐺𝑎𝑖𝑛, 𝑆𝑎𝑓𝑒𝐷𝑒𝑛𝑠𝑖𝑡𝑦) (1)
numerical models but most of them fail to adequately reduce
the routing overhead with highly scalable networks, which is a 𝑓(𝐿𝑜𝑤𝑒𝑟𝑅𝐴𝐷, 𝑈𝑝𝑝𝑒𝑟𝑅𝐴𝐷, 𝑃𝑟𝑜𝐷, 𝑀𝑖𝑛𝐺𝑎𝑖𝑛, 𝑆𝑎𝑓𝑒𝐷𝑒𝑛𝑠𝑖𝑡𝑦)
main feature of MANET. Those who focused on the DFCN = 𝑚𝑖𝑛 (𝑡𝑟 )
protocol did not formulate a trending mobility model for The function 𝑓 corresponds to the proposed system where
optimizing the decision parameters. Some of the researchers the target is to minimize the reachability time tr for each
directly focused on detecting the selfish nodes in the network instance of the simulator m. Table I below shows the DFCN
and avoiding them to increase the efficiency of the parameters along with their respective threshold and domain
broadcasting protocols, the most notable work in this regard is values.
by S. Subramaniyan et al. [16], where a Record-and-Trust-
223 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. DFCN PARAMETER DESCRIPTION TABLE. II. HIGHWAY MOBILITY SCENARIO PARAMETERS
As already stated, this will be done on three different Velocity [0.3 1.5] m/s-1
mobility model scenarios, namely, Highway, Mall and Human
mobility. The description for these scenarios is shown next. The human mobility scheme is defined as a round
simulation area, where fixed places that act as target spots are
A. Highway Scenario scattered and where the distance between two places cannot be
The main feature of the highway mobility model is that the less than 10 meters. Table IV shows the parameters for the
nodes move at significantly higher speeds compared to the human mobility model.
other mobility models and the nodes are lower in numbers. The
spot density is also set to one spot per square kilometer, which V. PROPOSED SYSTEM
is very sparse, and the number of spots per simulation area is The proposed technique consists of nested GA with fuzzy-
limited to three. In this scenario, most of the generated traffic based fitness. The aim is to optimize the DFCN decision
comes from nodes moving in opposite directions to simulate parameters according to the reachability time and to find
cars moving on different and opposing lanes of a highway. certain trends for each one of the different scenarios. The
Table II below shows the properties of this scenario. benchmark used is the reachability time for 10% of the nodes,
which is the time required so that 10% of the nodes in the
B. Mall Scenario network successfully deliver their messages. The outer GA
The mall mobility scenario is composed of separate regions contains the DFCN parameters and the to-be-calculated output
connected by relatively narrow areas. It represents a group of from the simulator. The inner GA evolves a set of rules for the
shops interconnected using corridors. In this scenario, the fuzzy system, where each chromosome represents a complete
surface area is smaller than the highway one and the velocity is fuzzy set and the inference output represents the inner fitness.
much slower. Also, the nodes move randomly for most of the The final inner fitness value that is calculated after the
time with no clear targets, representing humans wandering convergence has completed sets the fitness value of the outer
around and shopping in arbitrary shops. Table III illustrates the GA. The proposed system is developed using C# language on
different parameters for this scenario. Microsoft Visual Studio 2017 under 64-bit Windows 10 with
C. Human Mobility Scenario 8GB of RAM and an Intel Core i5-6500 CPU. Because the
proposed system is built using C# and the Madhoc simulator
This scenario is more distinctive than the mall one and is operates fully in Java, a mechanism that interfaces them was
considered one of the most daunting models. In this context, required. To be able to accomplish this, each time the simulator
the focus is on the human mobility scheme, where the is required to calculate the reachability time, it is executed by
movements are not random, but instead, there is a list of target the developed application as a command line program running
destinations that each node mostly moves towards. These inside a virtual sandbox process, where all the standard inputs
targets can be far away, as well as a few meters around. Also, and outputs are redirected to the application. Fig. 1 shows an
the targets can dynamically change with time depending on overview of the system.
human behavior. For instance, a waiter in a restaurant can be
regularly moving back and forth between the kitchen and Fig. 2 shows the pseudo-code of the proposed system.
customers’ tables.
224 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
225 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. V. NUMERICAL-TO-LINGUISTIC STRING CONVERSION TABLE trendline (red dotted line) is also calculated to provide a
mathematical model for the decision parameters. Fig. 5 shows
Value Equivalent Linguistic
the results for the highway mobility environment. Table VI
1 LOW shows the output trendline for each decision parameter and the
2 MED equivalent logarithmic regression expressions.
3 HIGH 5.2 9.5
-1 NOT LOW 5 9
UpperRAD (second)
LowerRAD (second)
-2 NOT MED 4.8 8.5
4.6 8
-3 NOT HIGH
4.4 7.5
0 NOT APPLICABLE 4.2 7
4 6.5
The selection is done through a traditional Roulette-Wheel 3.8 6
operator. It is worth noting that the last gene (reachability time) 3.6 5.5
is excluded from the evolution process and is stored inside the 3.4 5
chromosome and passed later to the fuzzy system. All of the
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
other aforementioned decision parameters are randomly Generations Generations
generated within the threshold. 72
0.65
C. Inner Genetic Algorithm 62 0.6
ProD (device)
The inner GA uses the same operators as the outer one.
MinGain
52 0.55
However, the chromosome structure is different. It consists of a
variable number of genes ranging from 3 to 15. Each gene 0.5
42
represents a key that encodes a linguistic string into numerical 0.45
values as shown previously. This had to be done in order to be 32
0.4
able to evolve the rules using the GA. Each key has a fixed
22 0.35
length of 6 which corresponds to the number of input
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
parameters and the output parameter.
Generations Generations
The population size for the inner GA is set to 50 and the 47 26.4
maximum number of generations is 100. Fig. 4 illustrates a 25.9
SafeDensity (device)
Time (second)
25.4
chromosome sizes, denoted with Sn, where n is the 37 24.9
chromosome number inside the population. It also shows an
example of how the key is decoded into a linguistic string. The 32 24.4
inner GA makes a complete run of 50 generations for each 23.9
27
outer chromosome. The target is to diversify the linguistics of 23.4
the fuzzy logic to reach the best possible output. 22 22.9
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
The defuzzified output value represents the fitness of the
outer chromosome. After doing this for all the outer GA Generations Generations
chromosomes, the best one is chosen and the outer GA transits 109
into the next generation. 108
107
Fitness
106
105
104
103
24
47
70
93
1
116
139
162
185
208
231
254
277
300
Generations
Fig. 5. Convergence for the High Way Mobility Model.
226 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 6 shows the results for the Mall mobility scenario and In the highway mobility scenario, the time to reach the
Table VII shows the trendline for the decision parameters. The destination decreased from 26.44 to 23.41 seconds, which
results for the human mobility model are shown in Fig. 7 and amounts to 11.45%. Given that the number of nodes in this
the respective trendline parameters are shown in Table VIII. network is 80, the average time for a node to deliver a message
decreased from 3.3 to 2.92 seconds.
5.2 7.9
6.5 9.4
UpperRAD (second)
LowerRAD (second)
UpperRAD (second)
LowerRAD (second)
4.2 6.9 5.5 8.4
5 7.9
3.7 6.4 4.5 7.4
3.2 5.9 4 6.9
3.5 6.4
2.7 5.4
3 5.9
2.2 4.9 2.5 5.4
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
2 4.9
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
Generations Generations
95 0.8 Generations Generations
0.75 37 0.6
90
0.7
ProD (device)
0.65 32 0.55
MinGain
85
ProD (second)
0.6
MinGain
80 0.55 27 0.5
0.5
75
0.45 22 0.45
70 0.4
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
17 0.4
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
Generations Generations
83 6.4 Generations Generations
78 72 4.1
5.9
SafeDensity (device)
73 4.05
Time (second)
62
SaeDensity (device)
5.4
68
TIme (second)
4
52
63 4.9
3.95
58 42
4.4 3.9
53 32
3.9 3.85
48
22 3.8
43 3.4
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
12 3.75
24
47
70
93
24
47
70
93
1
1
116
139
162
185
208
231
254
277
300
116
139
162
185
208
231
254
277
300
Generations Generations
502.9 Generations Generations
250
452.9
248
402.9
Fitness
246
352.9
Fitness
244
302.9
242
252.9
240
202.9
24
47
70
93
1
116
139
162
185
208
231
254
277
300
238
24
47
70
93
1
116
139
162
185
208
231
254
277
300
Generations
Generations
Fig. 6. Convergence for the Mall Mobility Model.
Fig. 7. Convergence for the Human Mobility Model.
TABLE. VII. TRENDLINE PARAMETERS FOR MALL SCENARIO
TABLE. VIII. TRENDLINE PARAMETERS FOR HUMAN MOBILITY SCENARIO
Parameter Trendline Expression
LowerRAD ↓ −0.463 ∗ 𝑙𝑛(𝑥) + 5.7409 Parameter Trendline Expression
UpperRAD ↓ −0.106 ∗ 𝑙𝑛(𝑥) + 7.3855 LowerRAD ↓ −0.088 ∗ 𝑙𝑛(𝑥) + 5.2098
ProD ↑ 0.9608 ∗ 𝑙𝑛(𝑥) + 82.223 UpperRAD ↑ 0.3832 ∗ 𝑙𝑛(𝑥) + 6.526
MinGain ↑ 0.0502 ∗ 𝑙𝑛(𝑥) + 0.445 ProD ↓ −0.757 ∗ 𝑙𝑛(𝑥) + 22.479
SafeDensity ↑ 2.0984 ∗ 𝑙𝑛(𝑥) + 60.488 MinGain ↓ −0.012 ∗ 𝑙𝑛(𝑥) + 0.5084
SafeDensity ↓ −7.621 ∗ 𝑙𝑛(𝑥) + 66.544
227 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
For the mall mobility scenario, the time to reach the nodes rigid optimization model, without the need to repeat the
decreased from 4.98 seconds to 3.57 seconds which amounts to evolution process each time.
28.3%, which brings down the average required time to deliver REFERENCES
a message from 7.6ms to 5.49ms.
[1] V. Rishiwal, S. K. Agarwal and M. Yadav, "Performance of AODV
As for the human mobility scenario, the time to deliver the protocol for H-MANETs," in International Conference on Advances in
Computing, Communication, & Automation (ICACCA) (Spring),
messages to their respective destinations decreased from 4.07 Dehradun, India, 2016.
to 3.78 seconds, which amounts to 7.12%. The average time to [2] L. J. G. Villalba, J. G. Matesanz, A. L. S. Orozco and J. D. M. Díaz,
deliver a message decreased from 0.5ms to 0.46ms. "Auto-Configuration Protocols in Mobile Ad Hoc Networks," Sensors
(Basel), vol. 11, no. 4, p. 3652–3666, 2011.
By inspecting all the previous results, it appears that the
[3] C. Dhakad and A. S. Bisen, "Efficient route selection by using link failure
mall mobility model benefited the most from the optimization factor in MANET," in International Conference on Electrical, Electronics,
of the DFCN decision parameters and the human mobility and Optimization Techniques (ICEEOT), Chennai, India, 2016.
model benefited the least. While these two models have very [4] P.-J. Chuang, P.-H. Yen and T.-Y. Chu, "Efficient Route Discovery and
close features, the major difference between them, as stated Repair in Mobile Ad-hoc Networks," in IEEE 26th International
previously, is the randomness of the movements. The human Conference on Advanced Information Networking and Applications,
mobility model is governed by human intentions of moving Fukuoka, Japan, 2012.
between a dynamic list of targets while the mall one is [5] V. Sharma and A. Vij, "Broadcasting methods in mobile ad-hoc
governed by random motion of shoppers moving between networks," in 2017 International Conference on Computing,
Communication and Automation (ICCCA), Greater Noida, India, 2017.
random shops. Also, by inspecting the highway scenario, it
[6] M. Bakhouya, "Broadcasting approaches for Mobile Ad hoc Networks,"
seems that the lack of enough nodes has significantly raised the in International Conference on High Performance Computing &
average delivery time six times (6x) the delivery time in other Simulation (HPCS), Helsinki, Finland, 2013.
scenarios. [7] H. Yadav and H. K. Pati, "A Survey on Selfish Node Detection in
MANET," in International Conference on Advances in Computing,
To demonstrate the consistency of the results, a 5% Communication Control and Networking (ICACCCN), Greater Noida
confidence interval for the final reachability time is calculated (UP), India, 2018.
and is shown in Table IX. [8] N. Ramya and S. Rathi, "Detection of selfish Nodes in MANET - a
survey," in International Conference on Advanced Computing and
TABLE. IX. 5% CONFIDENCE INTERVAL FOR THE FINAL REACHABILITY Communication Systems (ICACCS), Coimbatore, India, 2016.
TIME [9] L. Hogie, P. Bouvry, M. Seredynski and F. Guinand, "A Bandwidth-
Efficient Broadcasting Protocol for Mobile Multi-hop Ad hoc Networks,"
Mobility Model 5% Confidence Interval (seconds) in International Conference on Networking, International Conference on
Highway 23.4 ± 0.75 Systems and International Conference on Mobile Communications and
Mall 3.57 ± 0.34 Learning Technologies (ICNICONSMCL), Morne, Mauritius, 2006.
Human 3.78 ± 0.02 [10] B. Dorronsoro, P. Ruiz, G. Danoy, Y. Pigné and P. Bouvry,
"BROADCASTING PROTOCOL," in Evolutionary Algorithms for
Mobile Ad Hoc Networks, John Wiley & Sons, Inc, 2014, pp. 135-138.
VII. CONCLUSION AND FUTURE WORK
[11] L. Hogie and P. Bouvry, "An Overview of MANETs Simulation,"
The proposed system managed to decrease the message Electronic Notes in Theoretical Computer Science, vol. 150, no. 1, pp.
delivery time for the three real-life scenarios (the highway, the 81-101, 2006.
mall and the human mobility models) by optimizing the [12] L. Hogie, F. Guinand and P. Bouvry, The Madhoc metropolitan ad hoc
decision parameters for the DFCN protocol. The mall mobility network simulator, France: Luxembourg University and Le Havre
University, 2006.
model benefited the most from the optimization of the DFCN
parameters, which is mainly attributed to the randomness of the [13] L. Hogie, "Madhoc Metropolitan ad hoc network simulator," 2006.
[Online]. Available: http://www.i3s.unice.fr/~hogie/madhoc/. [Accessed
mobility, since the human mobility model also shares very 25 January 2019].
close parameters but only differs in the movement intention. In [14] R. M. Chintalapalli and V. R. Ananthula, "M-LionWhale: multi-objective
the human mobility model, the mobility is governed by the optimisation model for secure routing in mobilead-hocnetwork," IET
intentions of the humans to reach a certain dynamic list of Communications, vol. 12, no. 12, pp. 1406 - 1415, 2018.
destinations and, therefore, the randomness significantly [15] E. Alba, B. Dorronsoro, F. Luna and P. Bouvry, "A cellular multi-
decreases. Also, the highway mobility model yielded the objective genetic algorithm for optimal broadcasting strategy in
highest average message delivery time, which is attributed to metropolitan MANETs," in IEEE International Parallel and Distributed
Processing Symposium, Denver, CO, USA, 2005.
the lack of nodes and the very high mobility speed, and since
[16] S. Subramaniyan, W. Johnson and K. Subramaniyan, "A distributed
the DFCN protocol relies on 1-hop neighbors to deliver the framework for detecting selfish nodes in MANET using Record- and
messages to their destinations, this scenario severely affects it. Trust-Based Detection (RTBD) technique," EURASIP Journal on
Wireless Communications and Networkingvolume, p. Article 205, 2014.
In the future, a mathematical model based on the found
[17] S. S. Basurra, M. D. Vos, J. Padget, Y. Ji, T. Lewis and S. Armou,
trendlines can be established and tested. This will help to "Energy Efficient Zone based Routing Protocol for MANETs," Ad Hoc
achieve the results faster, instead of relying solely on Networks, vol. 25, pp. 16-37, 2015.
metaheuristic techniques, which require a significant amount of [18] M. Ahmad, A. Hameed, A. A. Ikram and I. Wahid, "State-of-the-Art
time to converge to the optimal solution. Clustering Schemes in Mobile Ad Hoc Networks: Objectives, Challenges,
and Future Directions," IEEE ACCESS, vol. 7, pp. 17067 - 17081, 2019.
Also, Genetic Programming (GP) can be experimented
with, to evolve programs and expressions related to each
scenario. This way, the resulting programs can be used as a
228 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Now-a-days Sindhi language is widely used in newspapers, Sindhi literature, books, educational/official
internet for the various purposes such as: newspapers, Sindhi websites, social network communication (What's App, Text
literature, books, educational/official websites and social messages, and social network), Teaching and learning
networks communications, teaching and learning processes. processes. In this regard, the use of the keyboard (Sindhi) is
Having developed technology of computer system, users face being increased day by day and on the other hand people are
difficulties and problems in writing Sindhi script. In this study, still facing the problem of unavailability of Sindhi keyboards.
various issues and challenges come in the Romanized Sindhi However, the communication system of local users is carried
text by using Roman transliteration (Sindhi text (ST) on by android based mobile phones services; these mobile
forms of Romanized Sindhi text) are identified. These
phones are unable to provide facilities to write Sindhi language
acknowledged issues are known as noise, written script of
Romanized and its style, space issues in Romanized script, some
containing 52 letters. Therefore, to overcome these problems,
characters not suitable in Romanized Sindhi, as a paragraph, Romanized Sindhi text is one of the best options [6].
rows, character issues, punctuation, row break and font style. Romanized Sindhi text is when used in different plate
However, this study provides the summary of issues and forms may face many issues and problems in writing of
challenges of Romanized Sindhi text. This research work Romanized Sindhi text or when use of different translators for
provides detailed information of issues and challenges faced by Sindhi of Romanized text. Also the use of translators and other
people during chatting in Romanized Sindhi text.
sources for normal users are very difficult and they need an
Keywords—Romanized Sindhi Text (RST); Sindhi language;
easy way for the solution of the problem.
issues and challenges; transliterator; social networks New issues and challenges of Sindhi language has been
communication found, when communicating with each other in Romanized
Sindhi text because Sindhi language has 52 letters of alphabets
I. INTRODUCTION
having different shapes, different symbols and different
Sindhi Language is a historical language of the world, the orientation of dots. So, it is very difficult to communicate using
majority of Sindhi language speakers are inhabited in Sindh Sindhi language on different social media. Therefore it is very
province of Pakistan. Around 12% peoples of Pakistan have important to have such platforms where people of different
mother tongue is Sindhi and an official language of the Sindh Sindhi community can communicate easily and properly using
[1]. Sindhi language is also spoken in different part of the Romanized format.
world with different ratio. Sindhi language has its own script
and written format. In Sindhi Language 52 alphabetical letters
(Fig. 1) were used for written as well as in speaking purposes
[2-4]. Since, Sindh language contains more alphabetical letters
than other languages, which causes difficulties for the new
learners. Sindhi script writing is a right handed script, same as
Arabic and Urdu Script. Urdu is morphological prosperous,
having different type of characters in Urdu script. Sindhi script
follows the rules as like Arabic Script and Perso-Arabic script
[5]. Fig. 1. Sindhi Language Alphabet.
In these days Sindhi language is considered as extensively
used in internet for the various purposes such as: daily
229 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
230 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
231 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
232 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
233 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Zeeshan Najam6
Department of Electrical Engineering
Ultimate Consultancy, Peshawar, Pakistan
Abstract—With the intention to meet the increasing demand bandwidth services (compared with P2MP) to be delivered to
of future higher bandwidth applications, fiber based Gigabit businesses or high rise buildings even over a long distance.
Passive Optical Network (GPON) access is considered best However, when the number of P2P connections is very high,
resolution to deliver triple play services (voice, data, video). the installation and maintenance costs can be prohibitively
Hence, it becomes obligatory to migrate from traditional copper- expensive.
based network to fiber-based. Due to rapid technological
evolution, tough competition and budget limitation the service On the other hand, the P2MP network based on GPON
providers are struggling to provide a cost effective solution to technology can provide an attractive solution to reduce the
minimize their operational cost with extra ordinary customer overall cost. With the P2MP GPONs, there are no electronic
satisfaction. One of the factors that increase the cost of overall components between an exchange and customer premises.
Fiber To The Home (FTTH) network is the unplanned Only optical splitters are used to connect Optical Line
deployment resulting in utilization of extra components and Terminal (OLT) equipment at an exchange to a group of
resources. Hence, it is imperative to determine a suitable premises sharing the same feeder fiber. An Optical Network
technique, which helps to reduce planning process, required time Unit (ONU) will then be used to convert the optical signal into
and deployment cost through optimization. Automation based an electronic signal at the customer’s premises.
planning is one of the possible ways to automate the network
design at probable lowest cost. In this research, a planning GPON has a downstream capacity of 2.488 Gb/s and an
technique for migration from copper to fiber access network with upstream capacity of 1.244 Gbp/s that is shared among users.
a manageable and optimized Passive Optic Network (PON – GPONs are generally considered to be a more cost effective
FTTx) infrastructure is presented identifying a cost-effective way of delivering FTTH services with minimum number of
strategy for developing countries. fibers and electronics required.
Keywords—Fiber To The Home; Passive Optical Networks; According to the market research division of Light
GPON; triple play; cost effective; customer satisfaction Reading [2], the number of households with fiber-optic
network connections was expected to grow by more than 32%
I. INTRODUCTION worldwide in 2009 and will continue to grow at rates close to
In order to provides a triple-play (voice, data, video) 45% a year through 2021. The number of fiber-connected
services by high speed collaborating apps, like games that runs households will reach nearly 130 million globally by that time.
online, several telecommunication organizations are Though, to form a cost effective GPON/FTTHN needs
considering Fiber based accessed networks as the key consideration of different factors, such as, locations of
resolution. There are two different ways of delivering fiber splitters, cable assignment of customer sites to splitters and
networks to customer premises, namely Point-to-Point (P2P) provision of spare capacity for future growth. In addition, all
and Point to Multi-Point (P2MP). On one hand, P2P [1] fiber the planning restraints like extreme allowable splitters
networks use a specific fiber constituent to link specific capability and the extreme distance in OLTs, ONUs, splitters
customer sites all the way to exchange. It allows very high satisfied must.
234 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
To plan a GPON/FTTH network manually in a new area, To The Home (FTTH) networks, FTTx Deployments Optical
an organizer is usually given a background plan that assisted CATV Networks, CWDM and DWDM Systems, Passive
by exchange. Whole region is subdivided in to small zones Optical Networks, Fiber Communication Systems Telecom,
that are settled in multiple phases. With given rules of LANs. It has the features like Low Insertion Loss, Ultra
planning and sites of locations, organizer typically positions
broadband performance (1260 –1630nm), Low PDL and
the optical Splitters (SPs) somewhere in the center of the
planned area. Cable Distribution points (CD)s will then be PMD, Stable towards thermal variations, Superior port to port
positioned around the SP afterwards. Once the locations of uniformity. A splitter type is shown in Fig. 2.
SPs and CDs are specified, the planner will assign cables from
each customer premise to a CD and from a CD to a SP based
on the shortest distance.
This manual design process is very time consuming. Very
often, due to the tight time schedule, when the design
proposed by the planner satisfies all the distance and capacity
constraints, he/she will submit the design without
incorporating much cost optimization or considering the
distribution of spare capacity.
A. Components of GPON FTTH Network
A Passive Optical Network (PON) is capable of having
P2M (point to multipoint) network with passive components
like optical splitter or coupler along the transmission section. Fig. 1. FTTH Network [3].
It uses active components only at CO and at customer
premises. It uses WDM to mix up video signals with the data TABLE. I. V8240 GPON OLT SPECIFICATIONS
and voice from OLT. Fig. 1 shows the basic FTTH Network. [3].
Flash Memory 72 MB
1) Optical Line Terminal (OLT): It is the most important SDRAM 1 GB
part of the network, where the electrical signal from the 17.1 x 12.2 x 11.2 in (434 x 310 x 285
Dimensions (W x H x D)
service provider’s equipment are converted into optical signals mm)
and given to the feeder network. The mode of transmission Switching Capacity 296Gbps
from ONT is broadcasting [4] from where it sends GEM Power Voltage AC type 100-240VAC, 50/60Hz
frames through the GEM port with GEM port IDs It is capable DC type -48/60VDC
of having Multi-service chassis for FTTx deployments, Operating Temp 32 to 122°F (0 to 50°C)
Supports a variety of service types, Non-blocking architecture SIU (Subscriber Interface Unit) 10 slots
NIU (Network Interface Unit) 2 slots
with & Routing within distributed architecture, scalability and
SFU (Switching Fabric Unit) 2 Slots
line rate performance, Full electrical and optical redundancy
Outstanding scalability and line rate performance, Real-time
TABLE. II. H640 SERIES GPON ONT
network traffic monitoring and analysis.V8240 GPON OLT is
used. Specifications are given in Table I. 4 10/100Base-TX ports (RJ45)
Service Interface 2 POTS ports (RJ11)
2) Optical Network Terminal (ONT): It is an active 1 RF video port (F-connector)
component used at customer premises which converts optical Uplink Interface 1 GPON port (SC/APC type)
to electrical signals. ONU/ONT represents the ingle customer Operating Temp 32 to 104°F (0 to 40°C)
where they will get the triple play application. H640 series Storage Temp -4 to 140°F (-20 to 60°C)
GPON ONT are used. It is capable of having carrier class Input 100-240VAC
VoIP telephony supporting both MGCP and SIP protocols, • Excluding bracket:10.24 x 2.05 x 7.87 in (260 x
Flexible VLAN tagging support, QoS for traffic prioritization 52 x 200 mm)
and bandwidth management, IGMP support for IPTV • Including bracket, wall mounting:10.51 x 2.60 x
Dimensions (W x H xD)
7.87 in (267 x 66 x 200 mm)
applications. Its specifications are given in Table II. • Excluding bracket, desktop mounting: 10.24 x
3) Splitter: Splitters are used to physically split the fiber 2.80 x 7.87 in (260 x 71 x 200 mm)
to number of fibers; to couple same or different information’s
to N users. MxN planar splitters are used which is based on
planar light wave circuit (PLC) technology and high precision
alignment. MxN splitters can split or combine light from one
or two fibers into N outgoing fibers uniformly over a wide
spectral range with ultra-low insertion loss and low
polarization dependent loss. With up to 64 output ports, these
splitters are ideal for high density split applications like Fiber
Fig. 2. PLC Splitter with Ribbon Fiber.
235 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
II. BUSINESS MODEL OF FTTH PLANNING control. Maintenance costs can also reduce because there
A. Business Model of FTTH Planning is no active device in the field to maintain, and optical
components have better reliability.
With the intention to meet the increasing demand of future
higher bandwidth applications, the fiber based access is Customer satisfaction will reduce the churn value and
considered to be a best resolution to offer triple play services. increase the customer, which also reduce operational cost.
It is therefore preferred with great need to migrate from To keep the existing customer is so easy as compare to
traditional capper based network to fiber based access. A enlist new one. To maintain an existing customer is so
business Model of new FTTH network deployment is
illustrated in Fig. 3 which consists of some dependent and easier than to register a fresh customer.
independent variable. Community benefits
In Pakistan the broadband growth in wireline is very slow FTTH enable Communities can get a lot of benefits with a
which is very much obvious. In [7], the slow growth is due to wider range of internet services. Few examples of possible
some factors that need improvement; these factors include. benefits with FTTH networks are as under:
Low literacy rate Financial boost with global competition.
low (level of) consumer awareness Attraction for new businesses.
Provisioning of state of the art services in Education and
No coverage of Broadband services
health sector.
Traffic reduction in broadband services low computer Improving overall quality of life in a community by
penetration
increasing the opportunities for communication.
Cost of service (tariff) Controlling of Road traffic blocking/problem.
History of market and national regulation
Business Model at a glance
1) Business output: The output of new FTTH Deployment
translates into different benefits [20]. Independent Variables Dependent variables
236 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
2) Automation based planning and process: To grip This Phase is especially problem particular as well as
complication of FTTHN, an automation scheme is settled by frequently modified to meet necessity. Moreover, the
keeping in mind, common optimization framework, which verification of this phase can be obtained by manual solution.
shows in Fig. 4 with different Phases. Calculation of Complete expenditure and design limitation
will validate in this phase. The business designer can validate
a) Input Phase: In this phase various sources can be their business model with the proposed Model. After the
used to retrieve the data like: completion of business model, it can be tested on various
Geo-graphical Information Systems (GIS) optimization methods.
Manually formed files d) Network Optimization Phase: This phase consists of
optimization approach based on Mixed Integer Linear
GIS data contains the setup of access network. Usually it
used a geographical database with a three-dimensional data Programming (MILP). It is a distinct variation of the linear
structure. To get the quick recovery of information, it programming. In MILP few variables have integer values. In
associates a wide range of geographic items with a rich set of MILP, our problem is molded by binary variables [8,11,15].
attributes. Prior to executing the MILP-based design tool, we
assumed that below given information’s are delivered:
Manually created files are the second source of input data
phase. A Map of an exchange or area is typically used by Locations of customer sites.
planner. The area further divided into different regions. The
planners usually choose a central point of the region for Location of one exchange E.
installation of Main Distribution Box (MDB) which consists Possible sites of cable distribution (CD).
of various splitters. Moreover, they allocate cables from end
user to splitters through distribution Cabinet (DC). Number of occupancies for all premises that
determines number of required PON links.
b) Input Analyzer Phase: As the data gathered by
different sources are in different format such as DXF, Esri Civil layer network which specifies connectivity in
Shape, so this stage is utilized to filter the required data from different network components.
Input source. The data is first transformed into some matrices Requirements of spare capacity needed to
which consist of required information to perform optimization accommodate future network growth.
procedures efficiently.
c) Business Logic Phase: The comprehensive cost III. MOTIVATION
model with engineering rules summarizes in this stage. To In this paper, a network design tool for the GPON/FTTH
reduce the cost of network design problems the engineering network is proposed to automate the planning process. Thus,
rules used for constrains. Different costs of network design given the locations of customer plots, possible locations of
like HR cost, ducting costs, cabling and network equipment CDs and SPs, the tool decides the optimal or near optimal
cost includes in this cost model. locations of CDs and SPs taking into account the future
growth and spare capacity distribution. In addition, cables are
Input Phase assigned from each plot to a CD and to the selected SP. The
Business Model Phase
solution is optimal in the sense that it leads to the minimum
Planning
cost of deployment which includes the number of network
Constrains Cost Model elements required, the total cabling distance and the
Rules
installation costs.
Input Analyzer
GIS Data Phase Optimization Phase
By using an automated planning assistant tool, the planner
Base can:
Minimize network capital expenditure, i.e., installation
Mixed Integer Linear Programming and materials costs.
Object
Model & Quickly achieve the network design of a given area.
Matrices
Manually Graph theory Related algorithm
Compare what-if scenarios to meet changes in planning
requirements.
Created
Data Rapidly re-cost networks for contract control and
Presentation Phase
installation.
237 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The benefits of automating, network design includes deployment. This algorithm initially discovers optimum
reducing installation and material expenses, decreasing time to number of clusters for customers and then continues to assign
make a design from hours to minutes, speedily re-costing every ONU to a cluster.
networks for different laboring or equipment costs and making
network design as well as automatically making bill-of- Researchers of [10] prolonged their research work in [11]
materials too. by producing multiple WDM PON networks concurrently. It
is attained by searching finest cost effective WDM PON by
IV. LITERATURE REVIEW splitting an area in to sub sections in which every sub region
covered by a specific PON. The authors of [12] proposed a
A. Literature Regarding FTTH Planning tool for semi-automated network planning. It defines a
The issue of FTTH network planning has been identified suboptimal route distribution for deployment cost. It utilizes
in literature. In this section two key methodologies are existing cable channels. After clustering customers, the
frequently used. The first one is that meta-heuristic techniques authors of this research utilize GA for route deployment
such as genetic algorithms (GA) and exact techniques like use process. The outcomes are compared to network designs
of Mixed Integer Linear programming (MILP) both attained by manual process which depicted that in most cases
methodologies have their pros and cons. The MILP can obtain this tool generates an inexpensive network.
global optimum solution as well as can outclass meta- Though, none of research work considered different types
heuristics for a medium-sized network [5]. of network elements selection, such as research work in paper
According to the authors of [6] presented a solution which [8] emphasized on decisive optimum locations of network
employs heuristics, falling in 2nd type of methodologies. Their elements and cable assigning. They assumed a specific type of
research emphasized on networks that are multistage splitting. network element. Another heuristic approach was recently
As they have complex problems, the researchers adopted introduced in [13]. The approach is based on clustering and a
heuristics for reducing computational time. The cluster based Tabu search and has been enhanced with mechanisms
and random locations of splitters were taken and performed handling resiliency issues as presented in [14]. The approach
comparisons. The effect of different localization of splitters on we use in this paper, originally presented in [15], is based on
lessening in CAPEX was deliberated. In [7] the authors beam search [16]. This approach has been enhanced with
proposed a model of a real life network. Real data of building mechanisms handling uncertainty issues following those used
sites and streets was mined from a Geographical Information in [17] and upgraded with the MIP polishing mechanisms of
System (GIS) by use of an open source map known as Open [18]. This last idea to mix MIP methodology with heuristics
Street Map (OSM). This model was adapted to single level proved to be very efficient and also has been recently used in
passive optical network (PON). These remote nodes are [19]. The methodology used in [20, 21] is MIP facilitated by
located for serving the customers. For more decrease costs the use of valid inequalities and various algorithmic
sustained, cables channels shared by different routes of cable. enhancements. Another recent work by Orange Labs is [22]. It
is similar to our research in majority of assumptions and the
In [8], the networks that based on MILP, a design tool for methodology used. However, it covers only the last access
GPON/FTTH networks are proposed. This tool automates the part of FTTH network; thus, the authors do not consider
planning process of networks. By providing positions of splitting and OLT costs. Still, the detailed view on the fiber
customers and probable positions of SPs and DC’s, it adopts splicing problem presented in [22] is definitely worth noting.
optimum positions of DCs and SPs as well as assignment of
cables in network elements and customers. Thus minimizes B. Literature Regarding GPON Technology
entire network development cost. The researchers took in to There are two main streams of research focusing on GPON
consideration future progression and introduced a technique technology: Dynamic bandwidth allocation (DBA) algorithms
for planning large networks. among OLT and ONUs, which can be found in [23, 24, 25]
According to [5], an approach is presented that is based on and optimal network design of the physical layer for GPON
meta-heuristic. This approach used Ant Colony Optimization deployment. The latter is the one considered in this research
(ACO). This algorithm achieves cable assignment on a and discussed in detail. Using the classical operational
multilevel network which emphasizing on cost minimizing. research approach, the planning problems can be assumed of
This algorithm allocates customers to DCs as well as DCs to as a ordered concentrator network problems. In context of
splitters simultaneously. Authors of [9] proposed a mixture of GPONs, the concentrator acts as a splitter to connect several
heuristics and mathematical programming for minimizing ONUs to an OLT in a star topology. When several splitters are
deployment cost of a GPON. The methodology of this connected to the OLT at different locations, it becomes a
research work was same to Simulated Annealing by which double-star topology. Details of the classical access network
assignment of cables as well as positions of splitters design approach can be found in [26, 27, 28, 29].
recursively reallocated till a well cost was found. In [10], a In [30], the authors developed an optimization solution to
scheme is presented that is based on previous class of perform multi-hierarchy PON planning. In their case, upper
methodologies. This scheme is a cross layer optimization. Optical Branching Devices (OBDs) and lower OBDs were
According to this scheme the researchers targets Greenfield introduced. The upper OBDs were used to connect between
network deployment and pursued to produce physical OLT and lower OBDs whilst lower OBDs were used to
architecture of Wave Length Division Multiplexing (WDM) connect between ONUs and upper OBDs.
PON networks. It attains nominal cost for network
238 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The locations of OBDs were calculated based on the Max- B. Decision Variables
Min Distance Cluster (MMDC) algorithm which can be found For optimization, the decision variables are followed:
in [10]. Regarding the optimization framework, authors in [31]
introduced a segmental framework which primarily Cabinet types utilization (given nodes)
emphasized on metaheuristic optimizing approaches. Practical
sample from motorized domain comprised for validate how All types of Splitter with locations utilization
overall problem could be break down in to sub-jobs and All types of cables utilization (given topology)
controlled through propose framework.
Splice closure and Splicing locations utilization
According to [32], the author works on collective
deployment of access network architectures such as Fiber To OLT types splice closures (locations are given)
The Node (FTTN), Fiber To The Micro Node (FTTN) and C. Problem Statement
Fiber To The Premise (FTTP) to decrease span of loops of
coppers through use of DSL access multiplexer in external Keeping in mind the structure as well as complexities of
cabinet and field micro node which are nearer to subscribers. complete problem, in our research description of its
Several classes of services and subscribers per class per point formulation is split in to four (4) problems that are depicted in
of demand are considered. The MILP model has been Fig. 5 as a square. We focused; these partial problems
proposed together with a tabular search base process for interlinked. The semantics of each variable that link to specific
improving computational time needed for finding best couples of partial problem is drawn by the shape of oval.
resolution. First part of problem is the bundle layer dimensioning
V. AUTOMATION LEADING TO OPTIMIZATION denoted by . Its purpose is that at assessing number with all
types of splitters that are installed and at selection number of
MIP approach is usually using in different means in FTTH OLT cards to be install at Central Office (CO) nodes as well.
networks designing. According to our research work, we adopt Delivered resolution promises that all accessing nodes,
MIP for the improvement of results that are returned by irrespective of its distance to the CO, delivered with requisite
empirical algorithms of optimization framework that are optical signal of appropriate power.
introduced in [33]. Framework used: locations of demand, the
available setup, with labor and equipment as well as The 2nd part of problem is cables which is denoted by
technology restraints. It returns a complete network planning . It defines number of cables of all types that are installed at
comprising: the topology of network, OLT, splices, splitters, every infrastructure segment. Installed the cables are to
OLT cards, cables, splices closures as well. By using stated provide fibers in numbers sustaining requests of bundle layer
aspects in modeling of a problem would leads to an incredible dimension problem.
of variables and restraints making acquired model unsolvable Splices are the 3rd part of problem. It is denoted by
by overall up-to-date MIP solvers. Hence, issue to be
which calculates number of optical closures and splices of all
shortened for making it amenable. Key supposition was for
types which are essentially be install in every infrastructure
using this approach is for improved acquired solutions;
node for supporting of solution that assessed in first two
simplifying of model not be depends on neglecting ostensibly
problems.
least significant factors, such as like splicing. While
apparently additional significant factors, such as the OLT The fourth part of problem is site dimensioning. It is
sites, to be static and detached from model. In this research, denoted by that’s goals is selecting a site type and number
optimization of capital expenses essential for the placement of with all types of hardware’s cabinet that are install in each
FTTH-OAN is addressed which contains one or more OLTs at infrastructure location.
the CO location and group of point of access that are located
in the / or nearby to the CP locations. Networks gratify loads
of all point of access take in to account permissible power
budget of optical links and split scenarios.
A. Prerequisites Data
The problem that is denoted by needs the given below
input data:
Passive and actives Equipment’s record.
Infrastructure networks Topology.
Every distribution and access node infrastructure paths
that are selected.
All access nodes Signal demand.
The Infrastructure sites that are decided for the
installing active, passive equipment. Fig. 5. Block Diagram of Proposed Structure for OLT.
239 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
240 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Infrastructure
Node
head end and access signal distribution points that are denoted N SLH
by , and repectively. This division
reveals level that engaged by specific splitters within network
links. The are real source of demanding signal demand Fig. 10. Signal Nodes with Links and Connections.
241 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Each head end signal distribution point n belongs to Following is categorization of bundle nodes:
essentially be allotted to single CO site , that is
a) Singleton set that represent by ≡{ } which
∈ , si_site(n)
masses the head end signal node . Each head end
Every access signal distribution point signaling distribution point ∈ , which is placed
can be allotted to any type of an in CO infrastructure location . Mentioning to sample
infrastructure site however CP, that is, si_site(n)∈ network single bundle head-node masses head end signal
, node and two head end signal distribution points
Lastly, every access signal node n that belongs to and .
essentially be allotted to any of CP infrastructure locations s b) Set of distribution bundle nodes. Every
that is belongs to distribution bundle node ∈ , agrees to CO or a DP
4) Model of bundle layer: The model of signaling layer infrastructure location ∈ and groups each
recognizes each distinct component must deliver signal distribution signal distribution points ∈ :
network links in OLT and ONT. In our work we presented si_site( ) = , positioned in that location. In sample
aggregate model, by the name of bundle layer. It deliberates network, there are two distribution bundle nodes ,
that collective, respectively, a subset of distribution signaling
groups of signal nodes and links as a substitute of individual
distribution points { , , ns } and { };–
ones.
set of accessing bundle nodes. Every accessing bundle
This model contains directed graph that is denoted node ∈ , related to an infrastructure location ∈
by ,, . The bundle nodes , and aggregates every access signal distribution point
denoted by while is group of bundle connections. ∈ : si_site( )= as well as each accessing
Organizes a contraction of a graph. signaling node ∈ : si_site( )= nis positioned
=( , of signaling layer, each bundle node at this location. In sample network as shown in Fig. 11, in
∈ signifies subset ⊆ of signaling nodes. All which three access bundle nodes represented by , ,
bundle layer links ∈ : ∈ δ( ) occurrence of respectively. We proposed c(n) : 2| |
bundle node signifies in turn group. function that for every distribution bundle node n ∈
distinguish subset {m ∈ : ∃l ∈ , b_a(l) = n ∧ b_b(l)
= { ∈ ∈ } of each signaling link = m} of bundle accessing nodes linked to that distribution
occurrence to this selected subset of signaling nodes. For node through a bundle connection; referred by distribution
simplify mapping in signaling and bundle layers model, we
cone of n distributing nodes.
present functions bs_nmap( | | . This
function expresses subset of signaling nodes amassed to c) Bundle links that denoted by further divided into
bundles node ∈ and function bs_lmap( trunk bundle and distribution bundle links that are represented
| | which states subset of signaling links denoted by a by and respectively. With support of bb_a(l) :
bundle link . and bb_b(l) : functions that
recognize, correspondingly, start and end bundle nodes of a
directed bundle link denoted by l ∈ , these two can be
properly define as, = {l ∈ : bb_a(l) ∈ , bb_b(l)
∈ } and = {l ∈ : bb_a(l) ∈ , bb_b(l) ∈
} respectively,. Sets and organize the
segregating of set .
The actual demands for signals network connections are
generates in accessing bundle nodes ; demand of
every distinct accessing bundle node n ∈ calculated by
given below expression:
∑ ∈ ∈ ∈ (2)
5) Concluding remarks of network fragment: The direct
bundle link l that is belongs to , of bundle layer maintained
through a group of similar fibers links in si_site(bb_a(l)) and
si_si te(bb_b(l)) infrastructure locations. By supposition, each
fiber link used identical infrastructure trail bi_p(l) ∈
where function bi_p(l) : expresses an
infrastructure trail that is taken by each fiber link supportive
Fig. 11. Bundle Nodes and Bundle Links. bundle link.
242 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Let directed trunk bundle connection from group . VII. CONCLUSION AND FUTURE WORK
Each fiber link which supports that link is known as trunk
fiber link. It uses a trunk infrastructure trail which contains The increasing demand for Broad band internet services
trunk fiber segments. All cable segments that have trunk fibers requires adaption of novel fiber base technologies. To attract
denoted as Trunk cable segment. Conferring to stated rules, new customers, fixed access network operators have to
we introduced group of trunk infrastructures trails substantially increase the speed and quality of internet
services. This can only be achieved by bringing the fiber as
representing by and group of distribution infrastructure
close to the customer as possible. This requires extensive
trails that represented by . We signify group of trunk
planning in term of cost, time and infrastructure. In this paper
infrastructure segment and distribution infrastructure segment we have presented a model of automating FTTH planning
through, respectively, ⊆ ⊆ ; we considering OAN. Different features, planning phases and
focused on sets that normally don’t organize model fragments have been identified and discussed, both
segregating of set . We assumed that each trunk theoretically as well as mathematically. In our future work we
infrastructure trail from which use single trunk will present formulation in term of optimization for FTTH
infrastructure segment from traverses segment in cost effective deployment. Real world experiment will aid in
identical direction. Therefore, trunk and distribution, formulation as well as validation.
infrastructure segment can be considered as directed. ii_a(l) :
ii_b(l) : , functions expresses REFERENCES
respectively, start and ending infrastructure nodes of [1] J. Prat, Next-generation FTTH passive optical networks: research
towards unlimited bandwidth access. Springer, 2008.[Online] Available:
infrastructure segments. The trunk segment in creates a http://dx.doi.org/10.1007/978-1-4020-8470-6.
directed tree by root at CO location , whereas distribution [2] PR Newswire,http://www.prnewswire.com/news-releases-62266647
segments in set creates forest of directed tree, that all .html,”April 2011.
rooted at a location which hosts distribution bundle node from [3] M.M.Al-Quzwini,―Design and Implementation of a Fiber To The
set. Hence, for every trunk segment l that belongs to Home FTTH Access Network based on GPON, in International Journal
there is just one or none predecessor trunk segment of Computer Applications, vol.92,no.6, April 2014.
ii_ah(l). Likewise, for every distribution segment k that [4] ITU-T G.984 Gigabit Passive Optical Network Specifications
belongs to there is just one or none predecessor [5] A. Chu, K. F. Poon, and A. Ouali, “Using Ant Colony Optimization to
distribution segment ii_ad(k). design GPON-FTTH networks with aggregating equipment,” in 2013
IEEE Symposium on Computational Intelligence for Communication
B. Equipment Catalogue Portion Systems and Networks (CIComms), April 2013, pp. 10–17.
[6] A. Eira, J. Pedro, and J. Pires, “Optimized Design of Multistage Passive
In this section with the help of catalog defining the Optical Networks,” IEEE/OSA Journal of Optical Communications and
physical equipment types are acceptable for installation at Networking, vol. 4, no. 5, pp.
infrastructure nodes and sites. Every catalog set is denoted by [7] O. Kipouridis, C. Machuca, A. Autenrieth, and K. Grobe, “Street aware
with a lowercase upper index; it is also used for individual infrastructure planning tool for Next Generation Optical Access
properties of an example of a particular type. Parameters networks,” in 2012 16th International Conference on Optical Network
common to every type, like cost or capacity, are denoted by Design and Modeling (ONDM), April 2012, pp. 1–6.
Greek letters and η with appropriate upper indices. A brief [8] K. F. Poon and A. Ouali, “A MILP based design tool for FTTH access
networks with consideration of demand growth,” in 2011 International
list of catalogue sets is listed in Table IV. The detail Conference for Internet Technology and Secured Transactions (ICITST),
description of this fragment is beyond the scope of this paper. Dec 2011, pp. 544–549.
[9] J. Li and G. Shen, “Cost Minimization Planning for Greenfield Passive
TABLE. IV. EQUIPMENT CATALOGUE SETS Optical Networks,” IEEE/OSA Journal of Optical Communications and
Networking, vol. 1, no. 1, pp. 17–29, June 2009.
Name Catalogue
[10] R. Chowdhury and B. Jaumard, “A cross layer optimization scheme for
Optical cables WDM PON network design and dimensioning,” in 2012 IEEE
OLT cards International Conference on Communications (ICC), June 2012, pp.
Cabinets 3110–3115.
Sites [11] “A p-center optimization scheme for the design and dimensioning of a
set of WDM PONs,” in 2012 IEEE Global Communications Conference
Segment preparations (GLOBECOM), Dec 2012, pp. 2977–2983.
OLT devices [12] B. Lakic and M. Hajduczenia, “On optimized Passive Optical Network
Fiber Splices (PON) deployment,” in Second International Conference on Access
Optical Splitter Networks Workshops, 2007. Access Nets ’07, Aug 2007, pp. 1–8.
Splitter Combinations [13] G. V. Arévalo, R. C. Hincapié, and R. Gaudino, “Optimization of
multiple PON deployment costs and comparison between GPON,
Joint Closures XGPON, NGPON2 and UDWDM PON,” Opt. Switching Netw., vol.
25, no. Supplement C, pp. 80–90, 2017.
243 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[14] G. V. Arévalo and R. Gaudino, “A techno-economic network planning [24] B. Skubic, J. Chen, J. Ahmed, L. Wosinska, and B. Mukherjee, “A
tool for PON deployment including protection strategies,” in 19th Int. comparison of dynamic bandwidth allocation for EPON, GPON, and
Conf. on Transparent Optical Networks (ICTON), July 2017, pp. 1–4. next-generation TDM PON,” IEEE Journals, vol. 47, pp. 40–48, 2009.
[15] R. Bisiani, “Beam search,” in Encyclopedia of Artificial Intelligence, [25] B. Chen, J. Chen, and S. He, “Efficient and fine scheduling algorithm
Wiley, 1987, pp. 56–58. for bandwidth allocation in ethernet passive optical networks,” IEEE J.
[16] M. Żotkiewicz and M. Mycek, “Impact of demand uncertainty models Sel. Topics Quantum Elect., vol. 12, no. 4, pp. 653 –660, 2006.
on FTTH network design,” in 18th Int. Conf. on Transparent Optical [26] C. Lee, “An algorithm for the design of multi-type concentrator
Networks (ICTON), July 2016, pp. 1–4. networks,” J. Oper. Res. Soc., vol. 44, pp. 471–482, 1993.
[17] Mycek M, Pióro M, Żotkiewicz M. MIP model for efficient [27] P. McGregor and D. Shen, “An algorithm for the access facility location
dimensioning of real-world FTTH trees. Telecommunication Systems. problem,” IEEE Trans. Commun., vol. 25, pp. 61–73, 1977.
2018 Jun 1: PP 1-20. [28] H. Pirkul and V. Nagarajan, “Locating concentrators in centralized
[18] F. D’Andreagiovanni, F. Mett, A. Nardin, and J. Pulaj, “Integrating LP- computer networks,” Annals of Operations Research, vol. 36, pp. 61–73,
guided variable fixing with MIP heuristics in the robust design of hybrid May 1992.
wired-wireless FTTx access networks,” Appl. Soft Comput., vol. 61, no. [29] S. Narasimhan and H. Pirkul, “Hierarchial concentrator location
Supplement C, pp. 1074–1087, 2017. problem,” Computer Communications, vol. 15, pp. 185–191, March
[19] Fritzsche, Lutz, Mathias Schweigel, and Rong Zhao. "Integrated 1992.
Network Planning: A Key Success Factor for Network Operators." [30] M. Lv and X. Chen, “Heuristic based multi-hierarchy passive optical
In Future Telco, pp. 43-52. Springer, Cham, 2019. network planning”in Wireless Communications, Networking and Mobile
[20] Hervet, C., Faye, A., Costa, M.C., Chardy, M., & Francfort, S. (2013). Computing, 2009.WiCom ’09. 5th International Conference on, sept.
Solving the two-stage robust FTTH network design problem under 2009, pp. 1–4.
demand uncertainty. In Proceedings of the international network [31] M. Lukasiewycz, M. Glaß, F. Reimann, and J. Teich, “Opt4J - A
optimization conference. Costa Adeje, Spain. Modular Framework for Meta-heuristic Optimization,” in Proceedings
[21] Angilella, Vincent, Matthieu Chardy, and Walid Ben-Ameur. "Fiber of the Genetic and Evolutionary Computing Conference (GECCO
cable network design in tree networks." European Journal of Operational 2011), Dublin, Ireland, 2011.
Research 269, no. 3 (2018): 1086-1106. [32] C. Steve, “Designing low cost access networks with iptv performance
[22] Angilella, V. (2018). Optimal design of Fiber To The Home constraints,” in Next Generation Internet Networks, 2008. NGI 2008,
networks (Doctoral dissertation, Institut National des 2008, pp. 45 –52.
Télécommunications) [33] Zotkiewicz, M., Mycek,M., & Tomaszewski, A. (2016). Profitable areas
[23] C. Bock, P. Chanclou, J. Finochietto, G. Franzl, M. Hajduczenia, T. in large-scale FTTH network optimization. Telecommunication
Koonen, P. Monteiro, F. Neri, J. Prat, and H. Silva, “Artchitecture of Systems, 61(3), 591–608.
future access networks,” in Next-Generation FTTH Passive Optical [34] ITU-T. (2001). Generic Functional Architecture of Transport Networks.
Networks: Research towards unlimited bandwidth access, 2008, pp. 5– Tech. rep. Recommendation G.805.
46.
244 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Social media has recently become a basic source for (known as “clickbait”) which implies non-credible content or
news consumption and sharing among millions of users. Social at least irrelevant content.
media platforms enable users to publish and share their own
generated content with little or no restrictions. However, this The extracted features are fed into different classification
gives an opportunity for the spread of inaccurate or misleading models which are then evaluated to identify the best
content, which can badly affect users’ beliefs and decisions. This performance given those set of features. The used techniques
is why credibility assessment of social media content has recently include: Logistic Regression (LR) [9]-[12], Decision Trees [1],
received tremendous attention. The majority of the studies in the [3], [6], [13]-[17], [19], Artificial Neural Networks
literature focused on identifying features that provide a high [20],[21],[22], Support Vector Machines(SVM) [6], [13], [14],
predictive power when fed to data mining models and select the [15], [17]-[21], [23] Random Forest (RF) [13], [15], [18], [24],
model with the highest predictive performance given those Naïve Bayesian (NB) [6], [16]-[19], [21] and K-nearest
features. Results of these studies are conflicting regarding the Neighbor (KNN) [17], [20], [21]. SVM and Decision Trees are
best model. Additionally, they disregarded the fact that real-time the most known and widely used models. Very few works tried
credibility assessment is needed and thus time and resources to use other models such as Linear Discriminant analysis
consumption is crucial for model selection. This study tries to fill (LDA) [21] and Adaptive Boosting (Adaboost) [23]. The
this gap by investigating the performance of different data performance of data mining techniques for credibility analysis
mining techniques for credibility assessments in terms of both
included only the most well – known techniques disregarding
functional and operational characteristics for a balanced
more advanced techniques that may better utilize the extracted
evaluation that considers both model performance and
interoperability.
features such as bagged and boosted ensemble models.
Moreover, the results of the performances are difficult to
Keywords—Data mining; performance evaluation; news compare as each study recommends a different model and
credibility; Twitter; social media therefore, no general agreement can be reached.
245 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
linear learning model, mixture discriminant analysis (MDA), TABLE. I. SUMMARY OF EMPIRICAL STUDIES OF DATA MINING MODELS
FOR CREDIBILITY ASSESSMENT
SVM, KNN, and NB. Both Multi-layer perceptron (MLP) and
learning vector quantization (LVQ) were selected as ANNs. Study Models Best performance
CART and C50 represent tree-based models and finally,
(SVM)
Bagging CART (BaggedCart), ADAboost, Gradient boosted Decision trees
machine (GBM) and RF represent ensemble learning models. [1] ERT
extremely randomized trees (ERT)
The selected models are evaluated based on accuracy, Naive bayes
precision, recall, F-measure and computational time
SVM
(processing and classification) and memory usage. [6] Decision trees Decision tree
This paper is organized as follows: in Section II, a review Bayes networks
of the previous empirical analysis of different data mining Decision trees
models in credibility assessment is presented. Section III [13] Random Forest Random Forest
provides a step-by-step description of the study methodology. SVM
Experimental results are discussed in Section IV. And Decision trees
[14] Decision tree
Section V concludes the study and sheds light on study SVM
limitations and possibilities for future work. Decision tree
[15] SVM SVM
II. DATA MINING FOR CREDIBILITY ASSESSMENT IN SOCIAL Random Forest
MEDIA Decision tree
[16] Decision tree
Data mining is a process that aims to analyze, identify Naïve Bayes
hidden patterns, and discover knowledge from large volumes SVM Decision tree for 1st
of data. Classification techniques are supervised techniques Naïve Bayes dataset
[17]
that classify data item into predetermined classes. These KNN KNN for the 2nd
decision trees dataset
techniques construct models using the labeled data to predict
the label of unknown data sets. Naïve Bayesian
[18] SVM Random Forest
The data mining process begins by applying data Random forest
preprocessing (i.e. data transformation, cleaning, feature Decision tree
selection, etc.) is applied to improve the classification [19] Naïve Bayesian Decision tree
efficiency of the algorithm. The data set contains each tuple is SVM
labeled to belong to a predefined class. Part of the tuples is Logistic Regression (LOG)
used for model construction (training dataset). The models are SVM
represented as classification rules or mathematical formulae [20] KNN Logistic Regression
and are tested using a set of independent data samples/tuples ANN
(test dataset) otherwise overfitting may occur. Finally, Decision trees
accuracy rate of the model is calculated as the percentage of ANN
test set tuples that are correctly classified by the model. Data KNN
mining techniques have been used for assessing the credibility [21] SVM ANN
of both information content and source. Credibility is assessed Naive Bayes
Linear discriminant analysis (LDA)
in terms of multiple features that are related to the news source,
content and propagation medium. Data mining techniques use SVM
[23] Adaboost
the features at one or more levels to label information content Adaboost
and/or source as credible/non-credible or fake/real. The
comparisons summarized in Table I were performed among III. METHODOLOGY
different models have conflicting results regarding their
A. Dataset
relative performance to one another. In the work [6], [19], DT
achieved higher performance than SVM while in [15] SVM The used dataset contains twitter news related to US
achieves better performance than DT. In [17], two different elections 2016. The dataset contains 9252 Twitter news
datasets were used and DT achieved the highest performance represented by 22 explanatory variables and one response
among other models given the first dataset while KNN was the variable. The predictors are related to both news content and
best given the second dataset. In [20] LR model outperformed source. The target variable labels each tweet to be fake/non-
more sophisticated non-linear models such as ANN, DT, and fake represented by (True/False) variable. The dataset contains
SVM. However, ANN proved higher performance in [21]. 254 instances labeled “unknown” and 2749 with no label. For
Ensemble models RF in [13], [18] and Adaboost in [23] proved this study unlabeled observations and noisy/unknown ones
higher performance over SVM. were disregarded.2 The remainder 5598 include approximately
87% labeled false/ to indicate non-fake/real news or other type
In conclusion, there is a need for a unified study that of news (i.e. comment, etc.), where 13% are labeled True to
analyzes the performance of different models and evaluates indicate fake news. Dataset metadata is presented in Table II.
their performance and applicability for credibility assessment.
2
Dealing with missing and noisy labels are out of the scope of this study.
246 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
247 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The variable with the highest VIF value is removed from 3) Variable selection: An important step before model
the dataset and the VIF test is repeated as values may change training is to select the features with the highest predictive
after each variable is removed. Results after removing power. For this study, features are evaluated and ranked based
“geo_coordinates” variable and repeating the test for the 2nd on the model in [27]. The model measures the effect of each
time indicated low VIF for “num_hashatgs” while both
variable on the target via an iterative variables' permutations
“user_followers_count” and “user_listed_count” still have high
VIF values as shown in Table III(b). process. The model calculates the mean decrease importance
of each variable based on which variable is confirmed or
The variable with the highest VIF value rejected. Results of the feature selection model confirmed all
“user_listed_count” was removed and the test was repeated. the selected variables as shown in Fig. 2 and Table IV.
Results of the 3rd test indicated low VIF value for all the
variables as shown in Table III(c).
248 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Variables VIF
retweet_count 1.013285
noofchars 1.366300
realname 1.467026
user_verified 1.833684
user_friends_count 1.058622
user_followers_count 5.362508
user_favourites_count 1.143870
geo_coordinates Inf
num_hashtags Inf
num_mentions 1.060546
num_urls 1.144046
Fig. 2. Mean Importance of the Explanatory Variables.
num_media 1.375882
user_listed_count 6.621056 TABLE. IV. MEAN IMPORTANCE OF THE EXPLANATORY VARIABLES
user_default_profile 1.239349
user_statuses_count 1.549332 Variable meanImp decision
Accountage 1.776116 user_statuses_count 23.14387 Confirmed
Isfake 1.004115 user_followers_count 20.36219 Confirmed
user_friends_count 19.07434 Confirmed
(B) VIF VALUES FOR THE 2ND TEST user_favourites_count 18.64027 Confirmed
accountage 16.55834 Confirmed
Variables VIF
user_verified 10.11282 Confirmed
retweet_count 1.014
realname 6.68721 Confirmed
Noofchars 1.360
user_default_profile 6.531671 Confirmed
Realname 1.456
num_media 5.164752 Confirmed
user_verified 1.852
noofchars 4.924822 Confirmed
user_friends_count 1.056
num_hashtags 3.645489 Confirmed
user_followers_count 5.435
num_mentions 3.544387 Confirmed
user_favourites_count 1.109
retweet_count 3.29834 Confirmed
num_hashtags 1.065
num_urls 2.544324 Confirmed
num_mentions 1.055
num_urls 1.148 C. Analytical Models
num_media 1.382 A set of the most known and most widely used models for
user_listed_count 6.689 fake news detection in the literature were selected for this
user_default_profile 1.228 study. The selected models cover different learning models
user_statuses_count 1.564 (linear, non-linear, tree-based and ensemble).
Accountage 1.774
1) Linea- learning models
Isfake 1.003
LDA: LDA is a linear learning model that tries to find
(C) VIF VALUES FOR THE 3RD TEST for a grouping of predictors that can discriminate two
targets. LDA is related to regression as they both try to
Variables VIF
retweet_count 1.013
express the relationship between one dependent
response variable and a set of independent variables.
Noofchars 1.364
However, LDA uses continuous independent variables
Realname 1.474
and a categorical dependent variable. The label for the
user_verified 1.815
new instance is estimated by the probability that inputs
user_friends_count 1.051 belong to each class and the instance is assigned the
user_followers_count 1.142 class with the highest probability calculated based on
user_favourites_count 1.088 Bayes Theorem [28].
num_hashtags 1.062
num_mentions 1.055 2) Non-linear learning models
num_urls 1.149 Mixture Discriminant Analysis (MDA): MDA is an
num_media 1.380 extension of LDA that models the within-group
user_default_profile 1.227 multivariate density of predictors through a mixture
user_statuses_count 1.233 (i.e., a weighted sum) of multivariate normal
Accountage 1.768 distributions [29]. In principle, this approach is useful
Isfake 1.005 for modeling multivariate non-normality or nonlinear
249 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
relationships among variables within each group, and multi-layer) and recurrent neural nets. Multilayer
allowing for more accurate classification. to determine perceptron (MLP) is a feed-forward ANN that contains at least
whether underlying subclasses may be present in each one hidden layer. Neurons in each layer use supervised
group. learning techniques [33]. LVQ is also a feed-forward ANN
SVM: A supervised learning model that analyses data that is based on the winner – takes – all learning approach. In
in order to identify patterns. Given a set of labeled this approach, the distance is measured between each data
training data, SVM represents instances in the dataset point and the output. The smaller distance indicates a winner
as points in a high-dimensional space and tries to which is then adopted by adjusting its weights. It’s as if, the
identify the best separating hyperplanes between prototype is moved closer if it correctly classifies the data
different classes. New instances are represented in the point or moved away if otherwise [34].
same space and are classified to a specific class based
4) Tree-based learning: Tree-based learning makes use of
on their closeness to the separating gap [30].
decision trees as a predictive model. Items are represented in a
NB: Naïve Bayesian (NB) is a classification technique tree structure. In such structure, nodes represent test points for
that is based on Bayes’ theorem [31]. It assumes variables, leaves represent class labels and branches represent
complete variables independence, as the a combination of variables that lead to class labels [35]. Two
presence/absence of one variable is unrelated to the popular implementations of DTs are a) CART [36] and C50
presence/absence of any other feature. It considers that [37]. CART is a binary DT that can be used for classification
all variables independently contribute to the probability
and regression. For classification, CART used Gini index
that the instance belongs to a certain class. NB bases its
predictions for new observations based on the analysis function to indicate the purity of the leaf nodes. C5.0
of previous observations. NB model usually outputs a algorithm is used to build decision tree or a rule set. It works
probability score and class membership. by splitting the sample based on the field that provides the
maximum information gain. It uses subsamples based on a
KNN: KNN is an Instance-based or memory-based variable and iteratively split data until subsamples cannot be
learning, labeling new instances is based on in-memory
split any further. Finally, the lowest-level splits are
instances stored in advance. In KNN, no internal model
is constructed, and computations are performed at reexamined, and those that do not contribute significantly to
classification time. KNN only stores instances of the the value of the model are removed/pruned.
training data in the features space and the class of an 5) Ensemble learning: Ensemble learning trains multiple
instance is determined based on the majority votes models using the same learning algorithm and set learners to
from its neighbors. The instance is labeled with the solve the problem. The main causes of error in learning are
class most common among its neighbors. KNN due to noise, bias, and variance. Ensemble minimizes these
determines neighbors based on distance using factors and may produce a more reliable classification than a
Euclidian, Manhattan or Murkowski distances for single classifier. Bagging (i.e. Bagging CART, Random
continuous variables and hamming for categorical Forest) and Boosting (i.e. Ada Boost and Stochastic Gradient
variables. Calculated distances are used to identify a
Boosting) get N learners by generating additional data in the
set of training instances (k) that are the closest to the
new point and label is assigned based on them [32]. training stage. N new training data sets are produced by
random sampling with replacement from the original set. By
3) ANNs: ANNs try to mimic the performance of the sampling with replacement, some observations may be
biological neural network of the human brain. ANNs are repeated in each new training data set. In the case of Bagging,
adaptive, fault tolerant and can learn by example. An ANN is any element has the same probability to appear in a new data
composed of a set of connected neurons organized in layers. set. However, for Boosting the observations are weighted and
The input layer communicates with one or more hidden layers, therefore some of them will take part in the new sets more
which in turn communicates with the output layer. Layers are often. Both are good at reducing variance but only Boosting
connected by weighted links. Those links carry signals tries to reduce bias, Bagging may solve the over-fitting
between neurons usually in the form of a real number. The problem, while Boosting can increase it [38].
output of each neuron is a function of the weighted sum of all
its inputs. The weights on the connection are adjusted during D. Performance Evaluation Metrics
the learning phase to represent the strengths of connections The performance of the selected models’ predictive power
between nodes. ANNs come with many structures. The most is evaluated based on accuracy, precision, recall, and F-
common structures are feed-forward neural network (single measure (F1).
250 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
251 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
works well only with categorical data and cannot perform on training, followed by MDA, SVM, and finally NB while MDA
continuous data. Thus, discretizing the continuous data may achieved the lowest classification time followed by KNN,
lead to better performance of this model. SVM, and NB. For memory usage, KNN had the minimum
usage during training and classification followed by MDA,
For ANNs – despite achieving 86.41% accuracy during NB, and finally SVM. For tree-based models, C50
training, LVQ accuracy dropped to 82.49% to achieve outperformed CART in training time while they both achieved
approximately 4% lower accuracy, 6% lower recall and 2% the same classification time. For memory usage, CART had the
lower F1 compared to MLP where precision of the two models lowest memory usage. For ANNs, LVQ outperformed MLP
is almost the same with 86.41% for MLP and 86.36% for LVQ. with lower time and memory usage in both phases. For
For tree-based learning models both CART and C50 trees ensemble learning models BaggedCart achieved the lowest
achieved the same performance over all metrics with 86.41% processing time while GBM achieved the lowest classification
accuracy, 86.41% precision, 100% recall and 0.9261 F1. For time and memory usage during both training and classification
ensemble learning – based models boosted models (GBM and among the rest of the models. It is worth noting that despite
AdaBoost) showed higher accuracy, recall, and F1 compared to their high processing time, AdaBoost achieved reasonable
bagged models (BaggedCart and RF) with 86.41% accuracy, classification time in relevance with the ensemble learning
100% recall, and 92.71% F1. However, BaggedCart achieved models. The best classification time was achieved by GBM and
86.59% precision which outperforms all the ensemble-learning LVQ (10 milliseconds), followed by CART, C50 and MDA
models. Comparison between different models is shown in (30 milliseconds). LDA had the lowest memory usage during
Fig. 4. classification, followed by KNN they both had less than 200
B. Operational Characteristics Evaluation. MB memory usage. In runtime, MLP and LVQ achieved the
Beside their predictive capabilities, operational lowest memory usage followed by KNN and LDA.
characteristics in terms of runtime and memory usage were A comparison between the models based on time and
tested for each model during both processing and classification memory usage is found in Fig. 5(a,b,c,d).
as shown in Table VII. the running statuses of each model was
obtained using “profvis” profiling tool in R. Results show Choosing the suitable model has to balance between high
variation in time and memory consumption as Adaboost has predictive performances, low classification time and memory
the maximum processing time which is much longer than all usage. That’s why LDA and CART can be recommended as
other models recording 1 hour 48 seconds and 350 they provide high predictive power with low time and memory
milliseconds while, the processing time of all other models usage compared to other models. GBM is recommended too as
ranged from 350 milliseconds for LDA (lowest processing it gives a good balance with the same performance with lower
time) to 42 seconds, 450 milliseconds for RF. For non-linear classification time and memory but higher processing time and
models, KNN achieved the lowest processing time during memory.
252 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Processing Classification
Model
Time(ms) Memory(MB) Time(ms) Memory(MB)
Linear Learning
LDA 00:00:00.350 104 00:00:00.090 2.4
Non-linear Learning
MDA 00:00:02.750 1419.4 00:00:00.030 3.7
SVM 00:00:14.550 4097.7 00:00:00.110 60.3
NB 00:00:20.650 2941.5 00:00:00.930 139.7
KNN 00:00:02.480 119.6 00:00:00.070 2.3
Tree-based Learning
CART 00:00:00.870 180 00:00:00.030 2.4
C50 00:00:08.750 690 00:00:00.030 3.1
ANN
MLP 00:00:12.580 14629.4 00:00:00.010 1.5
LVQ 00:00:06.150 355.3 00:00:00.010 1.4
Ensemble Learning
GBM 00:00:06.360 380.4 00:00:00.010 1.7
BaggedCART 00:00:05.880 1624.6 00:00:00.100 19.7
AdaBoost 01:00:48.530 34620.7 00:00:00.110 4.1
RF 00:00:42.450 4032.2 00:00:00.050 20.4
253 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
950
42500 900
39500 850
36500 800
750
33500 700
30500 650
27500 600
24500 550
500
21500 450
18500 400
15500 350
12500 300
250
9500 200
6500 150
3500 100
50
500
0
LDA
MLP
RF
C50
MDA
CART
LVQ
SVM
GBM
KNN
NB
BaggedCART
MLP
LDA
RF
C50
LVQ
SVM
MDA
KNN
CART
NB
BaggedCART
AdaBoost
GBM
(a) (b)
35000 140
30000 120
25000 100
20000 80
15000 60
10000 40
5000
20
0
0
LDA
RF
MLP
SVM
C50
LVQ
MDA
KNN
CART
AdaBoost
GBM
BaggedCART
NB
LDA
RF
MLP
SVM
C50
LVQ
MDA
CART
AdaBoost
KNN
GBM
BaggedCART
NB
(c) (d)
Fig. 5. (a) Processing Time of Models3. (b) Classification Time of Models. (c) Processing Memory usage. (d) Memory usage During Classification.
3
AdaBoost Processing time is not included due to its large value compared to other models (3648.53 second).
254 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
VI. CONCLUSION AND FUTURE WORK International Conference on Systems, Man, and Cybernetics, pages
2168–2172, 2011.
This study tries to present an evaluation of the [11] James Fairbanks, Natalie Fitch, Nathan Knauf, Erica Briscoe,
performances of different data mining models for credibility “Credibility Assessment in the News: Do we need to read?”, MIS2’18,
assessment in the context of social media. This study focused Feb 2018, Los Angeles, California USA.
on Twitter news credibility assessment as a case study. The [12] William Ferreira and Andreas Vlachos. “Emergent: a novel dataset for
bulk of works in the literature focused on identifying the most stance classification”. In Proceedings of the 2016 Conference of the
informative features, feed those features into different models North American Chapter of the Association for Computational
Linguistics: Human Language Technologies. 2016.
to select the model with higher predictive power and all of
[13] S. Kwon, M. Cha, K. Jung, W. Chen, and Y. Wang, "Prominent features
them disregarded time and memory consumption during both of rumor propagation in online social media", in Proceedings of
processing and runtime. Results of these studies contrast each International Conference on Data Mining, pp. 103-1108, 2013.
other and cannot give a unified decision. This study tries to [14] Zhe Zhao, Paul Resnick, and Qiaozhu Mei. “Enquiring minds: Early
address this limitation by benchmarking different data mining detection of rumors in social media from enquiry posts”. In Proceedings
models for news credibility assessment on Twitter. Models are of the 24th International Conference on World Wide Web . ACM, 1395–
evaluated in terms of their predictive performance using 1405, 2015.
Accuracy, Precision, Recall and F-measure and time and [15] Jing Ma, Wei Gao, Zhongyu Wei, Yueming Lu, and Kam-Fai Wong.
“Detect rumors using time series of social context information on
memory usage during both processing and prediction. microblogging websites”. In Proceedings of the 24th ACM International
However, the study still has some limitations and future on Conference on Information and Knowledge Management. ACM,
1751–1754. 2015.
research opportunities. First, the results on Twitter data may
[16] Aditi Gupta, Hemank Lamba, Ponnurangam Kumaraguru, and Anupam
not be applicable on different social media contexts (i.e. blogs, Joshi. “Faking sandy: characterizing and identifying fake images on
Facebook, etc.). One possible future research shall utilize twitter during hurricane sandy”. In Proceedings of the 22nd international
different datasets in different contexts for the evaluation. conference on World Wide Web. ACM, 729–736. 2013.
Another possible future work can be to explore the [17] Manish Gupta, Peixiang Zhao, Jiawei Han, “Evaluating Event
performance of other models including the less well known Credibility on Twitter”, Proceedings of the 2012 SIAM International
models and deep learning models. Performance can be Conference on Data Mining, pages = 153-164.
evaluated with missing and noisy labels. [18] Rim El Ballouli, Wassim El-Hajj, Ahmad Ghandour, Shady Elbassuoni,
Hazem Hajj and Khaled Shaban, “CAT: Credibility Analysis of Arabic
REFERENCES Content on Twitter”, Proceedings of The Third Arabic Natural Language
[1] Alexandra Olteanu, Stanislav Peshterliev, Xin Liu, Karl Aberer, “Web Processing Workshop (WANLP), pages 62–71, 2017.
Credibility: Features Exploration and Credibility Prediction”, in the [19] Sahar. F. Sabbeh, S. Batawah, Arabic news credibility on Twitter: An
proceedings of European Conference on Information Retrieval. ECIR Enhanced Model using Hybrid Features", Journal Of Theoretical And
2013:Advances in Information Retrieval pp 557-568, 2013. Applied Information Technology , Vol 96 April 2018.
[2] John ODonovan, Byungkyu Kang, Greg Meyer, Tobias Hollerer, Sibel [20] Michela Del Vicario, Walter Quattrociocchi, Antonio Scala, Fabiana
Adal, “Credibility in Context: An Analysis of Feature Distributions in Zollo, “Polarization And Fake News: Early Warning Of Potential
Twitter”, In the proceedings of the International Conference on Social Misinformation Targets”, Arxiv:1802.01400v1 [Cs.Si] 5 Feb 2018.
Computing and 2012 ASE/IEEE International Conference on Privacy, [21] R.Deepa Lakshmi , N.Radha , “Supervised Learning Approach for Spam
Security, Risk and Trust, 2012. Classification Analysis using Data Mining Tools “, (IJCSE)
[3] A. A. A. Mansour, "Labeling Agreement Level and Classification International Journal on Computer Science and Engineering, Vol. 02,
Accuracy," 2016 12th International Conference on Signal-Image No. 09, p= 2783-2789, 2010.
Technology & Internet-Based Systems (SITIS), Naples, 2016, pp. 271- [22] Marin Vuković ,Krešimir Pripužić, Hrvoje Belani, “An Intelligent
274. doi: 10.1109/SITIS.2016.51 Automatic Hoax Detection System”, In Knowledge-Based and
[4] Dana Movshovitz-Attias, Yair Movshovitz-Attias,Peter Steenkiste, Intelligent Information and Engineering Systems , pages 318–325.
Christos Faloutsos, “Analysis of the Reputation System and User Springer, Berlin, Heidelberg, September 2009.
Contributions on a Question Answering Website: StackOverflow”. In [23] Benjamin Markines, Ciro CaŠuto, and Filippo Menczer, “Social spam
the Proceedings of the IEEE/ACM International Conference on detection. In Proceedings of the 5th International Workshop on
Advances in Social Networks Analysis and Mining Pages 886-893. Adversarial Information Retrieval on the Web . ACM, 41–48, 2009.
2013.
[24] Martin Potthast, Johannes Kiesel, Kevin Reinartz, Janek Bevendorff,
[5] Ruohan Li, Ayoung Suh , “Factors Influencing Information credibility Benno Stein” A Stylometric Inquiry into Hyperpartisan and Fake
on Social Media Platforms: Evidence from Facebook Pages”, In the News”, arXiv:1702.05638, 2017.
proceedings of the 3rd Information Systems International Conference
(ISICO2015), 2015. [25] James Gareth, Witten Daniela, Hastie Trevor, Tibshirani Robert, “ An
Introduction to Statistical Learning (8th ed.)”. Springer
[6] C. Castillo, M. Mendoza, and B. Poblete, “Information credibility on Science+Business Media New York. ISBN 978-1-4614-7138-7, 2017.
twitter”, In the Proceedings of the 20th international conference on
World wide web, Hyderabad, India, 2011. [26] Dietterich, T. G. “Approximate Statistical Tests for Comparing
Supervised Classification Learning Algorithms”. Neural Comput,
[7] Meredith Ringel Morris, Scott Counts, Asta Roseway, Aaron Hoff, Julia 10(7):1895–1923, 1998.
Schwarz, “Tweeting is Believing? Understanding Microblog Credibility
Perceptions”, CSCW 2012, USA. [27] Miron B. Kursa, Aleksander Jankowski, Witold R. Rudnicki, "Boruta –
A System for Feature Selection", Fundamental Informaticae volume101,
[8] Kanda Runapongsa Saikaew, Chaluemwut Noyunsan, “Features for pages:271–285, 2010.
Measuring Credibility on Facebook. Information”. In the proceedings of
the XIII International Conference on Computer Science and Information [28] McLachlan, G. J. Discriminant Analysis and Statistical Pattern
Technology (ICCSIT 2015),Thailand, 2015. Recognition. Wiley Interscience. ISBN 0-471-69115-1. MR 1190469.
2004.
[9] Eugenio Tacchini, Gabriele Ballarin, Marco L. Della Vedova, Stefano
Moret, and Luca de Alfaro, “Some Like it Hoax: Automated Fake News [29] Fraley, C., & Raftery, A. E. Model-based clustering, discriminant
Detection in Social Networks”, CoRR,abs/1704.07506, 2017. analysis, and density estimation. Journal of the American Statistical
Association, 97(458), 611-631.2002.
[10] Mehrbod Sharifi, Eugene Fink, and Jaime G. Carbonell. “Detection of
Internet scam using logistic regression”. Proceedings of IEEE
255 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[30] Cortes, Corinna; Vapnik, Vladimir N. "Support-vector networks". [38] Rokach, L. (2010). "Ensemble-based classifiers". Artificial Intelligence
Machine Learning. Volume: 20 No:3: p:273–297. Review. 33 (1-2): 1–39.
doi:10.1007/BF00994018, 1995.
[31] Russell, Stuart; Norvig, Peter. Artificial Intelligence: A Modern AUTHOR’S PROFILE
Approach (2nd ed.). Prentice Hall. ISBN 978-0137903955. 2003. SAHAR F. SABBEH earned her B.Sc. degree
[32] Altman, N. S. (1992). "An introduction to kernel and nearest-neighbor in information systems from the Faculty of
nonparametric regression". The American Statistician. 46 (3): 175–185. computers and information technology, Mansoura
doi:10.1080/00031305.1992.10475879. university, Egypt in 2003. She earned her M.Sc.
also in information systems from the same
[33] Rumelhart, David E., Geoffrey E. Hinton, and R. J. Williams. "Learning department in 2008 and earned her Ph.D. degree in
Internal Representations by Error Propagation". David E. Rumelhart, 2011 She has been a member of the IEEE since
James L. McClelland, and the PDP research group. (editors); 2017. Dr. Sabbeh worked in Alzarka High
[34] Parallel distributed processing: Explorations in the microstructure of Institution for management information systems
cognition, Volume 1: Foundation. MIT Press, 1986. from 2004 to 2009. She worked at Misr Higher
[35] T. Kohonen, "Learning vector quantization", in M.A. Arbib, The Institution of Engineering and Technology, Mansoura, Egypt from 2009 till
Handbook of Brain Theory and Neural Networks, Cambridge, MA: MIT 2011. She worked with the Faculty of Computers and Information
Press, pp. 537–540, 1995. Technology, Banha University, Egypt during the period from 2011 - 2018 as
an assistant professor. She also worked part time as an assistant professor in
[36] Rokach, Lior; Maimon, O. “Data mining with decision trees: theory and several reputable private universities in Cairo, Egypt. She worked as an
applications”. World Scientific Pub Co Inc. ISBN 978-9812771711. Associate professor in the faculty of computers and information technology,
2008. King Abdul-Aziz university, KSA during the period from 2016–2018.
[37] Breiman, Leo; Friedman, J. H.; Olshen, R. A.; Stone, C. J., Currently, she is an associate professor at the Faculty of Computers and
“Classification and regression trees”. Monterey, CA: Wadsworth & Information Technology, Banha University, Egypt and an associate professor
Brooks/Cole Advanced Books & Software. ISBN 978-0-412-04841-8. in the computer science and engineering, university of Jeddah, KSA. She
1984. supervised 5 M.Sc. and one Ph.D. students.
256 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The e-commerce emotion analysis is notable and dreary procedure and may create upsets in basic decision
the most pivotal advance since it catches the customer emotion in making [4].
a product, and emotions with respect to product to decide if the
customer attitude is negative, positive, or neutral. Posting on the In this way, an effective method for showing the general
customer's reviews have turned into an undeniably famous path emotion of a product dependent on customer reviews is
for individuals to share with different customers their emotion required. This paper inquire about in the investigation of
and feelings toward a product. This review has a significant product reviews are worried about ordering the general
impact on sales in the future. The proposed system utilizes mixed emotion for a specific product. As a customer review does not
word from an adjective (adj) and adverb to improve the emotion have a standard structure and may incorporate spelling
analysis process utilized a rule-based emotion analysis. The blunders and equivalent words for the product features,
system extracts an Arabic customer review and computes the emotion classification per feature can be troublesome.
frequency of each word. At that point, it computes the emotion
and score of each customer review. The system likewise computes Emotion AI is a procedure of extracting information from
the emotion and score of straightforward Arabic sentence. users' assessments. The decisions of the people get influenced
by the conclusions of other individuals. Today, if any person
Keywords—Component; rule-based; emotion; customer review; needs to purchase an item then the person will initially look
Arabic through the surveys and emotion about that item via an online
shopping, a social network like Twitter, Facebook, and other
I. INTRODUCTION user forums, at that point recognizable proof of assumption,
With the rapid development of web applications, social turns out to be extremely troublesome from this colossal
network and online shopping, there moved toward becoming information physically. Thus, there is a need for a
audits, comments and feedback generated by customers. These computerized emotion analysis system. The fundamental goal
emotions can be about essentially anything, including products, of this paper is to perform emotion AI for Arabic sentences.
politics issues, news and service. All of which should be
handled and broke down to get a good estimation of what the II. RELATED WORK
customer thinks and feels. Before the accessibility of automatic Elhawary and Elfeky [5] utilized that gathering Arabic
emotion analysis tools, the way toward getting customer business reviews, and dedicating 80% of the gathered business
surveys was an incredibly cumbersome and time-consuming reviews to prepare their classifier which is utilized to recognize
task [1]. review's records. They developed various Arabic vocabularies
used to investigate distinctive Arabic reviews and emotion. The
Numerous emotion AI were created for English, yet in this
extremity of every Arabic business review whether it is:
paper are attempting to break another ground in this field and
positive, negative, neutral or mixed is made a decision about
concoct a high accuracy Arabic based emotion analysis tool
dependent on the assembled dictionaries.
which isn't influenced by the utilization of vernaculars; a tool
that enables Arab customers to analyze the e-commerce Diverse strategies were utilized by El-Halees [6] to decide
shopping, enabling them to know the general feeling about the extremity of various Arabic `s. The extremity of the entire
products being talked about. The Arabic language has Arabic comment is resolved first utilizing the vocabulary-based
numerous lingos that ought to be considered, wherein every technique, where the output from the primary strategy
vernacular implications of words can be very surprising. (dictionary based) is considered as a preparation set for greatest
Arabic is a morphologically rich language and this can raise entropy strategy, which is utilized to order these comments.
issues for any programmed content examination instrument [2,
3]. Another methodology has been proposed depends on
translating the source Arabic emotions into English and after
The enormous increment in e-commerce shopping in that utilization the equivalent relevant procedures to examine
middle east especial Egypt, gulf countries, that made customer the came about English emotions. Almas and Ahmad in [7]
reviews significant in decision making procedure of a utilized machine translation systems to translate the source
customer. The quantity of reviews for a product can be very comment or review from Arabic to English language before
high, particularly for a most prevalent product. A significant passing them to an English based emotion analysis system. The
number of customers is interested on emotion of a product, so issue of this methodology was the loss of nuance after
for this reason, they should initially read all the reviews to translating the source to English.
reach a resolution. What's more, since perusing countless is a
257 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Rushdi-Saleh et al. [8] utilized another methodology was B. Procedure of Final Emotion and Score
machine learning algorithm to arrange the extremity of Arabic Most of the work use adjectives only for emotion analysis,
reviews extricated from specific Web pages identified with and some of them use nouns, verbs, adverbs or a combination
motion movies. Inui et al. [9] think about receive making an of them. This proposed model use emotion adj and nouns of
interpretation of suppositions from English to Japanese, trailed adj words because that some emotion do exclude any adj,
by emotion analysis. They applied sentiment-oriented sentence however, express a negative or positive sentiment such as; The
filtering strategy to alleviate numerous interpretation mistakes vast majority of the work use adjectives just for emotion
that happen as a reaction of interpretation to decrease the investigation, and some of them use noun, verb, adverbs or a
impact of interpretation blunders in multilingual comment level mix of them. utilize emotion adjectives and noun of adjectives
review. since they noticed that some emotion statements do exclude
Choi et al. [10] presents a structure for emotion analysis, any adjectives, however, express negative or positive emotion,
focus around the feeling piece of information that is identified for example, “” رائع, “”جامد جدا, “” ممتاس, “” جميل جدا, “” حلو مزة.
with a supposition theme, for example, company or individual. That means “Wonderful”, “Excellent”, “Very beautiful”, “Very
They utilize a domain-specific sentiment classifier for every nice”. All of them are adjectives in this example, although it
domain with the recently totaled signs (for example a subject expresses positive emotions.
or the theme of the emotion) in light of a proposed semi-
supervised strategy. Yi et al. [11], Kim et al. [12], Choi et al. IV. APPLYING RULE
[10] extricate emotion about a subject spotlight on the An algorithm is proposed to extend and recognize the
estimation piece of information that is identified with a emotion AI automatic of new feeling words utilizing total
conclusion theme. This is characterized as an essential subject procedure and free online Arabic word references and
of supposition articulation in a sentence, for example, vocabularies and calculate the frequency of each emotion word
organization, individual or occasion. from a dataset. The scores of adj are appointed between -1 and
1. In the event that any of the token matches with terms of adj
III. MATERIAL AND METHODS and adverb score content then their comparing scores are
To perform emotion AI, basic Arabic content record, tweets processed. At that point, the last emotion is assigned to content
or comment in online shopping are inputted by the client. At as positive, negative or natural utilizing calculation has been
that point, the system takes a shot at it and figures its emotion proposed below. Be that as it may, if any of the tokens in the
and score. The design appeared in Fig. 1 show the working of content does not match with terms of adj score. At that point, 0
rule-based emotion AI system. scores are assigned to that content lastly no emotion is assigned
to that content. For example: “”هذا المنتج جميل جدا. In this
A. Tokenizer example, the process of this content will be split into tokens as
The system takes a product review in the Arabic language “”هذا, “” المنتج, “”جميل, “”جدا. Presently, these tokens are
as an input; the input sentence is part into tokens through coordinated with adj and adverb score content. Here, “” جميل,
tokenizer. A token is a piece of an arrangement of characters in “ ”جداare found in adverb and adj word score content
content that are combined together as an important semantic separately. Along these, relating scores of “”جميل, “ ”جداare
unit for handling. The tokenizer changes over a sentence into appointed as 0.1 and 1 individually. And afterward, the last
word level tokens comprising of a word, accentuation marks, score and emotion are registered by a system utilizing the
and different symbols. proposed model in this paper.
A. Role of Score Adjective and Adverb
Product
Review The score document contains the rundown of pre-processed
emotion scores as appeared in the table below. Each line of the
scoring document contains a word or expression alongside its
emotion score. In the event that a word or expression which is
Tokenization found in content yet not found in score document at that point
word or expression is given an emotion score of 0.
At that point, the emotion of content based on assumption
scores of the terms in the content is registered. The emotion of
Yes No content is equivalent to the whole of the assumption scores for
Found each term in the content.
adj
Grammatical forms information is most generally used in
matching linguistic tasks. It is used to disambiguate sense which
Assign = 0
subsequently is used to coordinate component decision [13].
Researchers basically use adjective (adj) words and adverb as
highlights to discover the emotion in content. Adjective (adj)
Assign emotion
words are most normally utilized as highlights among all
Fig. 1. Proposed Architecture of Rule-based Emotion AI System.
grammatical features. There is a solid connection amongst adj
and subjectivity of content. Indeed, even every one of the
grammatical features assumes a critical role, yet just adj words
258 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
as highlights feature the emotions with high exactness. An V. DESCRIPTION OF THE PROPOSED ALGORITHM
exactness of around 82.8% has been accomplished in film First step is to take an Arabic sentence as output for a
survey spaces by utilizing adj words just as highlights [14]. A system, the second step is the identification of relations and
few instances of positive and negatives have appeared in widespread words, third step is assigning a score to emotion
Table I. according to pre-processed emotion scores as appeared in table
B. Role of Score Content previous, fourth step calculation a final score according to
degree of strong or weak function according to relation
1) Calculate strong function: If the score of an adj is identification as shown below.
greater than 0 then adj is positive and adv has a place with
positive. For example, “ ”جميلhas meant in English "beautiful" If score > 0
is a positive adj and “ ”جداmean in English "Very" has a place
with a strong adverb. Then Arabic sentence show positive emotion,
2) Calculate weak function: If the score of an adj is less
than 0 then is a negative emotion and adverb has a place with else if score < 0:
is negative then score of both adj and adverb will be less than
0. For example, “ ”سئthat means in English “bad” is negative then the Arabic sentence show negative emotion,
adj and “ ”جداthat mean in English “very” is also a negative
adverb. The model has been proposed to calculate strong and else: the Arabic sentence shows no emotion or normal
weak functions in this work as shown in Fig. 2. emotion.
259 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[7] Y. Almas,K. Ahmad, “A note on extracting „sentiments‟ in financial International CIKM Workshop on Topic-Sentiment Analysis for Mass
news in English, Arabic & Urdu,” In: The Second Workshop on Opinion Measurement(TSA'09), Hong Kong – China, pp. 37-44, 2009.
Computation, al Approaches to Arabic Script-based Languages, [11] S-K. Kim, E. Hovy, “Determining the sentiment of opinions,” In:
Linguistic Society of America 2007 Linguistic Institute, Stanford Proceedings of the 20th international conference on computational
University, Stanford, California., Linguistic Society of America, pp. 1– linguistics (COLING 2004), Geneva, Switzerland. pp. 1367–1373, 2004.
12, 2007.
[12] M. Elhawary,M. Elfeky, “Mining Arabic Business Reviews,” In:
[8] M. Rushdi-Saleh, M. Teresa Martín-Valdivia, L. Alfonso UreñaLópez,J. Proceedings of the2010 IEEE International Conference on Data Mining
M. Perea-Ortega, “Bilingual Experiments with an ArabicEnglish Corpus Workshops; pp. 1108-1113, 2010.
for Opinion Mining. Language,” In: Proceedings of Recent Advances in
Natural Language Processing, Hissar, Bulgaria. pp. 740-745, 2011. [13] Pang B, Lee L 2008 Opinion mining and sentiment analysis.
Foundations and trends in information retrieval, 2(1-2):1-135.
[9] T. Inui, M. Yamamoto “Applying Sentiment-oriented Sentence Filtering
to Multilingual Review Classification,” In: Proceedings of the [14] Pang B, Lee L, Vaithyanathan S 2002 Thumbs up? Sentiment
classification using.
Workshop on Sentiment Analysis where AI meets Psychology (SAAIP),
IJCNLP, Chiang Mai, Thailand, pp. 51–58, 2011. [15] Machine learning techniques. Proc. ACL-02 Conf. on Empirical
methods in natural language processing, 10: 79-8.
[10] Y. Choi, Y. Kim,S-H. Myaeng, “Domain-specific Sentiment Analysis
using Contextual Feature Generation,” In: Proceedings of the 1st
260 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The Maritime Transport industry in the statistical report compiled by the Philippine Maritime Industry
Philippines has been growing through the years and has been a Authority (2016), there is an increasing rate of passenger
catalyst in the industrial development of the country. Although traffic yearly [3]. In 2017, a total of 72,438,609 passenger
the maritime transport sector is one of the largest industries in traffic based on the total embarking and disembarking data
the country, the safety devices and technology used are sluggish was reported. Alongside the volume of passengers are risks
phase to change. The natural hazards and human error are main regarding their safety. The Philippines being situated near the
cause of maritime incidents, resulting to multiple casualties and Pacific Ocean makes it one of the most vulnerable countries in
missing persons every year of which this study seek to address the world to weather-related extreme events. The Philippines
the problem of safety in the maritime transport industry. The
placed second among 171 countries ranked on their risk level
study aims to design and develop a system that will locate an
overboard1 passenger whenever a vessel is in distress. The
to disasters the report added that the country lacks 80.03% of
Floating Overboard Accident Tracking System (FLOATS) was coping capacity to minimize the negative consequences of
conceptualized by combining the Search Theory, Theory of natural hazards and climate change through direct action and
Planned Behavior (TPB) and Disaster Preparedness, and the the resources available [4]. In 2012, there were 610 reported
increasing availability of tracking device and monitoring persons killed or missing (lives lost) worldwide [5]. In
technologies and the advancement of communication systems. addition to this, it was also stated that human errors and
The system consists of the Global Positioning System (GPS) for fatigue were featured eminently in these accidents. Over the
location data, Global System for Mobile (GSM) communications recent years, most casualties recorded from maritime incidents
for the transmission and reception of emergency messages, are from passenger or roll-on/roll-off (RO-RO) ships and
Arduino-Nano microcontroller to handle the processing, the used general cargo ships. There are 185 average number of deaths
of an inflatable life jacket with signal light and a rescue update due to maritime accidents in the country yearly [6]. In addition
display using an organic light emitting diode (OLED) for the to this, a report released by MARINA in 2016 states that there
search and rescue operations. Tests and surveys established the were 707 Search and Rescue (SAR) missions, 211 casualties,
functionality, reliability, and acceptability of the system, which and 216 persons missing related to maritime incidents on that
will greatly benefit maritime incident responders by securing year. Several memorandum and circulars were made to
vessel passengers from hazards and reducing the time allotted increase maritime safety and resilience to hazards [7]. An
through speedy search and rescue operations. example of a provision under a memorandum is to upgrade the
maritime safety infrastructures like the navigation aids,
Keywords—Global Positioning System (GPS); Global System
for Mobile communications (GSM); Organic Light Emitting Diode
lighthouses, vessel monitoring services and systems, and other
(OLED); Arduino-Nano microcontroller; tracking system; life maritime ancillary services like the weather bureaus.
jacket; life jacket light Memorandums and circulars were also made to prevent
maritime accidents. While strict regulations are implemented,
I. INTRODUCTION the number of search and rescue operations as stated earlier in
Maritime transport is the foundation of globalization and is a report by MARINA in 2016 is relatively higher compared to
the center of cross-border networks of transport that aid 126 number of maritime search and rescue operations in 2015.
supply chains and empower international trade [1]. The Moreover, the increasing number of passenger traffic yearly,
Philippines having composed of 7,641 islands makes its implies the need for an appropriate technology to counter the
shipping transport industry a vital part of the economic growth consequences of maritime incidents and promote disaster
and development. One may assume that because of the resilience particularly in the maritime transportation sector.
archipelagic nature of the Philippines, the maritime industry is This study provides the design and development an
traditionally anchored in its economy [2]. The country’s appropriate technology to be used in times of maritime
archipelagic setting requires an efficient maritime transport incidents. The technology is known as Floating Overboard
infrastructure and systems of safety aids. According to a Accident Tracking System (FLOATS) was integrated in a life
* Corresponding Author jacket with an extend battery life span that sustains the
261 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
tracking device1 through a solar panel therefore assuring a 4) OLED Display5 – Organic Light Emitting Diode
higher survivability, reducing the exposure of passengers (OLED) Display is an efficient and thinner display that can
involved in a maritime incident/accident or natural hazards light up individual pixels when necessary and to different
and minimizing the time allocated in locating the strayed degrees. The maximum resolution of the OLED Display is
passengers aiding the authorities in search and rescue mission.
128x64. The OLED display shows the rescue updates sent by
Next section describes the system design model of FLOATS.
the search and rescue authorities.
II. SYSTEM DESIGN 5) Ni-MH Battery6 – the battery used in the tracking
Fig. 1 shows design model of the FLOATS prototype device has a voltage of 7.2V with a capacity of 2200 mAh.
components hardware and software developing the tracking 6) Solar Panel7 – used in addition to mains-supply
device that is integrated to the life jacket, increasing the chargers for energy saving during the daytime with a
probability of success of a search and rescue operation maximum voltage of 17.5 V, current of 0.57A and maximum
through GSM, GPS, and OLED output. power of 10W.
B. Software
Arduino Integrated Development Environment8 (IDE) –
Arduino IDE is an open-source software program that allows
user to write and upload code within a real-time work
environment. The IDE had been instrumental in writing,
compiling, and uploading codes to the Arduino Board. Writing
the individual codes for each module and integrating all of
them were vital in developing the firmware for the tracking
device.
III. METHODOLOGY
Fig. 2 illustrates the comprehensive flow of the whole
research process to guide the researchers in creating a
prototype with an appropriate technology namely the Floating
Overboard Accident Tracking System (FLOATS).
A. Analysis and Data Gathering
In this stage, the researchers started gathering data and
relevant information, conducting preliminary investigation and
interviews. Gathering and supporting facts about the existing
problem to be able to design and develop the FLOATS. The
Fig. 1. Design Model of the FLOATS. researchers investigated first the scale of the problem in the
Philippines. Having read articles and releases by the World
A. Hardware Risk Report of the United Nations University Institute for
Environment and Human Security, the researchers found out
1) Arduino Nano Boar2 – Arduino Nano is a small, that the Philippines placed second among the 171 countries
complete, and breadboard-friendly board based on the ranked on their risk level to disasters. The Philippine lacks
ATmega328P. It has 22 Digital I/O pins and 8 Analog IN pins. eighty-percent 80% of coping capacity to minimize the
The microcontroller board is used to handle the processes negative consequences. With the increasing number of
needed by the tracking device. passenger traffic in the maritime transport industry of the
2) SIM800L GSM/GPRS Modul3 – SIM800L is a GSM country, the researchers saw the importance of safety in this
Module that features a complete Quad-band GSM/GPRS particular sector. Secondary data from the Philippine Maritime
solution in a LGA type. The module can transmit Voice, SMS, Industry Authority (MARINA) and the Philippine Coast
and data information with low power consumption. The Guard (PCG) further supports the claim that there is an
increasing number of search and rescue operations regarding
tracking device’s SMS interfacing is made possible by
maritime incidents in the Philippines. Researchers conducted a
SIM800L. survey at the City Disaster Risk Reduction Management
3) Neo M8N GPS Module4 – Neo M8N is a GPS Module Office and Philippine Coast Guard Iligan Station concerning
used to retrieve location and time information from GPS the existing tracking devices [12]. Based on the interviews
satellites. The Neo M8N GPS Module is a significant conducted with the authorities, the researchers found out that
component of the tracking device that receives location there were no tracking devices used to locate or track
information of the overboard passenger. overboard passengers when doing search and rescue
operations.
6
1
https://www.merriam-webster.com https://en.wikipedia.org/wiki/Nickel%E2%80%93metal_hydride_battery
7
2
https://store.arduino.cc/usa/arduino-nano https://www.conserve-energy-future.com
8
3
https://lastminuteengineers.com/sim800l-gsm-module-arduino-tutorial/ https://en.wikipedia.org/wiki/Arduino_IDE
4
https://www.u-blox.com/en/product/neo-m8-series
5
https://www.thefreedictionary.com/OLED
262 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
263 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
264 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The researchers also considered the behavioral concepts of imply that by using the FLOATS tracking device, in a matter
a passenger bound to be exposed to hazards caused by a of seconds whenever vessel passengers are forced overboard,
maritime incident. The Theory of Planned Behavior stated that rescuers can already receive location data and start tracking
attitude, social pressure, and control are the factors that the passengers.
determine an intention to engage in a behavior at a specific
time and place [9]. As to the problem concerning this study, This would significantly decrease the time allotted for
the researchers considered the place where a person engages a searches and uplift the feeling of assurance for the vessel
specific behavior which is in a maritime incident setting. passengers. The researchers conducted a real-time test of the
When the FLOATS is implemented during a maritime tracking device on the surface waters of the Panguil Bay. Two
incident, the attitude of a vessel passenger that is bound to be members of the research team were on the field to test the
distressed might change from a state of apprehension to a state tracking device. Meanwhile, the third member was assigned to
of assurance. Social pressure might inflict panic among vessel gather the location data sent by the tracking device. The third
passengers. However, the system might reduce the level of it. member stays in the Disaster Risk Reduction Management
Office (DRRMO) of the Municipality of Tubod, Lanao del
The study emphasized the development and integration of Norte. The two members on the field were accompanied by
the tracking device with solar panel to the life jacket. After the two rescuers from Tubod-DRRMO to guarantee the safety of
components were individually tested and integrated to form the testing. In every location point, the boat used by the
the tracking device, the problem on whether the tracking research team will turn to a halt until 5 location data are
device was possible to be fully-integrated to the life jacket obtained. The trial lasted for two hours. The areas included in
surfaced. Fig. 6 shows the final integration of the prototype. the testing of the tracking device include the municipalities of
Tubod and Baroy in the province of Lanao del Norte and
Since the researchers have used prototyping modules to Barangay Silanga of Tangub City. Table III exhibits all the
make the tracking device possible, it was difficult to fully location points gathered during the testing of the tracking
integrate the tracking device to the life jacket. The researchers
device.
view the size of these prototyping modules as the main
contributory factor to the obstacle of the integration. In Trial 1 of the first location had the location result of
addition to this, altering the original design of the life jacket 8.078998 and 123.795753. The first set of numbers,
might dispute its original function and damage it in the long specifically 8.078998, indicates the latitude coordinate of the
run. To deal with this obstacle, the researchers used a location data. The latter indicates the longitude coordinate.
carabiner and attached the tracking device to the strap of the Both are vital for its plotting on a web-based mapping service.
life jacket. By using a web-based mapping service called Google Maps
and its counterpart Google Earth, the researchers plotted all
The functionality of the FLOATS is evaluated by the location data resulting to the image (Fig. 7).
recording the results throughout the whole system operation
specifically the response time of the device. Table II exhibits
the response time of the tracking device SMS receive from
each trial.
In the overall response time test result, five trials were
executed by the researchers. The time was tracked using a
smart phone’s stopwatch. Through the results shown on the
table, each operation time (in seconds) is averaged. It would
take an average of 4.698 seconds for the initialization message
to be displayed. On the other hand, the SMS sent display
would take an average of 9.798 seconds. From the passenger’s
location, it would take an average of 8.288 seconds to reach
the rescuer from the time the SMS sent display is flashed on
the tracking device. And, the rescue update sent by a rescuer
to the passenger in average would take 3.98 seconds to be
displayed on the OLED of the tracking device. The results Fig. 6. Integration of the Prototype.
Location Message
No. of Trials Initialization Display (sec) SMS Sent Display Rescue Update Display (sec)
Receive (,) (Sec)
Trial 1 4.82 9.58 8.57 5.41
Trial 2 3.44 9.74 7.76 3.22
Trial 3 4.90 10.09 8.38 4.40
Trial 4 4.81 9.82 8.68 3.30
Trial 5 5.52 9.76 8.05 3.57
Average 4.698 9.798 ok 8.288 3.98
265 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
No. of Trials
Location Area
(Latitude, Longitude) Trial 1 Trial 2 Trial 3 Trial 4 Trial 5
Tangueguiron, Tubod
(Somewhere
8.078998 , 123.796287 8.078836 , 123.795768 8.078648 , 123.795799 8.078311 , 123.795799 8.078311 , 123.795799
Alim Shrine)
8.080128 , 123.796287
Tubod Port
8.056834 , 123.782714 8.055989 , 123.782196 8.055922 , 123.782096 8.055735 , 123.781997 8.055574 , 123.781898
8.05708 , 123.782714
Sagadan, Tubod
(MCC HOTEL) 8.047289 , 123.77414 8.045537 , 123.772735 8.04344 , 123.772544 8.045242 , 123.772415 8.044876 , 123.772117
8.048721 , 123.775749
Baroy, Tubod
(Seaside Cuzina Bar) 8.033945 , 123.75846 8.033936 , 123.758201 8.033894 , 123.758102 8.033854 , 123.757987 8.033809 , 123.757843
8.033973 , 123.758476
Silanga, Tangub City
(Silanga Port) 8.064497 , 123.775749 8.064688 , 123.776054 8.064755 , 123.776222 8.064462 , 123.776809 8.064395 , 123.777114
8.064160 , 123.77534
Fig. 7. Google Maps Location Data Result. Fig. 8. Location Data Result.
266 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Strongly Strongly
Statement Disagree Neutral Agree
Disagree Agree
TOTAL 8 54 118
267 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
mobile network and its signal strength. The acceptability part [4] United Nations University Institute for Environment and Human
of survey results show that the tracking system was met with Security (UNU-EHS) (2014). The World Risk Report 2014. Available:
online. https://i.unu.edu.
high acceptance with 66.29% strongly agree the use of
[5] Stipe Galic, Zvonimir Lusic, and Ivica Skoko (2014, April 28-29). 6th
FLOATS. International Maritime Science Conference (IMSC): The Role and
This study effectively develops a tracking device with the Importance of Safety in Maritime Transportation. Availble in
https://bib.irb.hr
appropriate technology that can be used by passengers during
[6] Martinez RE, Go JJ and Guevarra J (2016). Epidemiology of drowning
vessel capsizing at sea. The researchers emphasize the deaths in the Philippines, 1980 to 2011. Western Pac Surveill Response
importance of safety in the maritime transportation industry J. 2016 Nov 8;7(4). doi:10.5365/wpsar.2016.7.2.005. Available: online.
and by applying the system to the current situation of the http://ojs.wpro.who.int
industry, the risks regarding passenger safety while at sea can [7] Orlando S. Dimailig, Jae-Yong Jeong, and Chol-Seung Kim (June
be reduced. 2011). Marine Transportation in the Philippines: The Maritime
Accidents and their Causes. Available: online.
ACKNOWLEDGMENT https://www.researchgate.net
[8] U.S. Coast Guard Research and Development Center. Review in Search
This Research work is done by the undergraduate student Theory: Advances and Applications to Search and Rescue Decision
of Bachelor of Science in Electronics and Communication Support. Report No. CG-D-12-01, Washington ,DC. National Technical
Technology with the advice and support of the faculty of Information Service, SpringField, VA 22161. Available: online.
Computer Application Department. https://apps.dtic.mil/dtic/tr/fulltext/u2/a397065.pdf
[9] Icek Ajzen (1991). The theory of planned behavior. Organizational
REFERENCES behavior and human decision processes 50 (2), 179-211
[1] United Nations Conference on Trade and Development (UNCTAD) [10] Arwa Masoud Hamza El-Nasri (June 2011). Design and
(2016). Review of Maritime Transport: Fostering the transparency of Implementations of GPS Mobile Tracking System: Overall Tracking
maritime markets and analyzing relevant developments. Retrieved May Centre Design. Retrieved December 12, 2018, from
23, 2018, from https://unctad.org http://khartoumspace.uofk.edu
[2] Lara Richter (2016). The Impact of the Maritime Industry on the [11] Ralph H. Balingasa , Maria Tricia Camille R. Bilog , Jonnelle Klenn D.
Philippine Economy. German-Philippine Chamber of Commerce and Castillo , Jerome M. Perez , Agnes F. Terrible , & Rionel B. Caldo.
Industry, Inc. Makati City 1234, Philippines. Avilable in Distress Signal Tracker Using GPS and SMS Technology: A Prototype.
https://philippinen.ahk.de/fileadmin/AHK_Philippinen/Publications/Mar Retrieved December 12, 2018, from http://lpulaguna.edu.ph
itime_Industry_in_the_Philippines__GPCCI_2016_.pdf
[12] Ernesto Empig, Joel Miano, Harreez Villaruz, Nieva Mapula, et.al,
[3] Maritime Industry Authority (MARINA) (August 2018). Annual Report 2015, Development of Digital Human Body Tracker Alarm System
on Basic Maritime Statistics: the available maritime and maritime- Using GPS and Transceiver for Catastrophic Events Rescue Operation
related statistical information from the year 2012 up to the year 2016. (DHBT AS), 8th AUN/SEED-Net Regional Conference on Electrical
Retrieved October 8, 2018, from http://marina.gov.ph and Electronics Engineering. https://uyr.uy.edu.mm/handle/
123456789/371.
268 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Student retention is an important issue faced by programs. According to Seidman [8], the “formula for student
Philippine higher education institutions. It is a key concern that success is: Retention = Early Identification and Early Intensive
needs to be addressed for the reason that the knowledge they Continuous Intervention”.
gain can contribute to the economic and community development
of the country aside from financial stability and employability. Educational data mining (EDM) can be used to resolve this
University databases contain substantial information that can be student retention problem. EDM „refers to a method for
queried for knowledge discovery that will aid the retention of extracting information from large collection of data in
students. This work aims to analyze factors associated with educational institutions through data mining (DM) techniques
student’s success among first-year students through feature to extract useful knowledge to help decision makers‟ [12].
selection. This is a critical step prior to modelling in data mining, Records of students can be queried as an attribute dataset, such
as a way to reduce computational process and improve as admission test scores and socio-demographic attributes.
prediction performance. In this work, filter methods are applied These can be utilized as predictors for the prediction model for
on datasets queried from university database. To demonstrate knowledge discovery in databases (KDD).
the applicability of this method as a pre-processing step prior to
data modelling, predictive model is built using the selected The two of the three most popular model used in extracting
dominant features. The accuracy result jumps to 92.09%. Also, knowledge from data are KDD process model (shown in
through feature selection technique, it was revealed that post- Fig. 1) and Cross-Industry Standard Process for Data Mining
admission variables are the dominant predictors. Recognizing (CRISP-DM) model [9]. Both models contain data
these factors, the university could improve their intervention preprocessing phase, which is crucial and tedious. In fact,
programs to help students retain and succeed. This only shows performing the tasks in this phase can consume considerable
that doing feature selection is an important step that should be amount of time. This includes data cleaning, data
done prior to designing any predictive model. transformation, and data reduction. An overview of common
DM preprocessing steps will be discussed in details in the
Keywords—Educational data mining; feature selection; data succeeding section.
preprocessing; knowledge discovery; student retention
However, this paper will only use filter selection feature
I. INTRODUCTION methods: Correlation Feature Selection, Information Gain
Universities have continuously experience challenges in Ratio, and Chi-Square analysis. To sought if the results of
retaining students. Accordingly, about 40% of students in these selection methods will vary, it will be tabulated and
tertiary will not graduate on time [1]. This has been a pressing ranked according to feature importance, and will be compared.
problem in universities around the world. As „higher education This study also aims to cite evidence in support of feature
enrolments have increased in recent decades, dropping out of selection method as part of preprocessing step to increase the
university has become a common experience‟ [2]. Like in the classification accuracy of a predictive model which has been
Philippines, Commission on Higher Education (CHED) omitted in some DM studies; like in the following similar
records show that there has been a 4.1 million to 3.6 million studies [10],[11], and [12]. In view of this, two predictive
total number of dropped out students between academic year models using classification technique with different feature
2015-2016 and 2016-2017 [3]. Further, according to the datasets is proposed– model 1 will used all the dataset
survey, “only 23% of Filipinos finish college” [4]. attributes queried from the university database and model 2
Undergraduate college enrolments have grown increasingly but will used the ranking of important features. Moreover, feature
with less graduates. Yet, few researchers in Philippine- selection method in this study is utilized to identify the possible
educational-community have addressed attrition and retention factors instrumental to student retention and as part of data
problems. reduction phase. The significance of this result affects the
First-year is regarded in this study considering that it has student and society, along with financial consequences for the
high attrition rates [5]. It has been affirmed in the study of institution.
Garett, Bridgewater and Feinstein [6] that first year is vital in The structure of this paper is as follows: Section 2 reviews
indicating academic success and considered very important at some similar works of this study and presents feature selection
many educational institutions [7]. Thus, the assistance and methods used in this study. Then in Section 3, presents the
monitoring of first-year students should be regarded because
universities can respond to these students through intervention
The author has been supported by the Commission on Higher Education,
Individual Research Grants (CHED M.O. No. 14-2016).
269 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
methodology while Section 4 discusses the results. Finally, pre-academic data and GPA. The model JRip rule induction
Section 5 provides conclusion and future work. has the highest accuracy result of 77.30%. Data transformation
and Data cleaning were the only pre-processing steps made.
270 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
X2
o e
2
e (2)
where o is observed frequencies and e is expected
frequencies. This method is used to identify whether a Fig. 3. Methodology Flow.
distribution of observed frequencies varies from the supposed
expected frequencies. To remove the irrelevant data and noise from the dataset,
the following steps were carried out.
E. Informaton Gain Ratio (IGR)
1) Load data collected
IGR method computes the importance of the features using
2) Integrate collected data
information gain and give weights to them accordingly even if
it applied to features that have dissimilar value using the 3) Filter data by removing missing records
equation below [18]. 4) Remove duplicates
5) Do normalization
IG(att )
GR(att ) 6) Detect Outlier
H (att ) (3)
Careful data integration is done to reduce and avoid
where equation (4) redundancies and inconsistencies. Redundant data were
carefully examined; same attributes were not included in this
H (att ) j P(vj ) log 2 P(vj ) study.
(4)
where P (vj ) corresponds to the chances of having vj by Data cleansing steps is performed to remove the incomplete
data. A list-wise deletion method is adopted to delete the entire
providing general values for an attribute j. record from the analysis if any variable in the model has a
III. METHODOLOGY missing value. Missing data is ignored to avoid adding bias and
distortion to the dataset. Removing a few records will not
Fig. 3 illustrates the activity in this study. In the data pre- impede the results of the model since this study contains large
processing feature selection is emphasized. dataset. Finally, to handle outliers, local outlier factor (LOF) is
A. Dataset Collected executed.
The records used in this study were real records of five C. Feature Selection
academic years queried from a university database. These One of the main goals in this study is to identify what
records contain information about the entrance result, grades, dominant variable or combination of variables collected can be
and among others. The data for this research was inputted in a used as predictors of first year student success. In this study,
data mining tool. The dataset is comprised of 7, 936 records filter model using feature rankings are used, namely, the Info
with 29 features. Gain Ratio, the Correlation Feature Selection, and the Chi
The potential predictor variables queried fall into two Square, to identify the dominant variables. The significance of
categories: pre-college data and post-admission data. Pre- using filter model method is that it separates feature selection
college data are records prior to admission, it includes from learning [19]. Thus, no bias towards any learning
admission test scores and socio- demographic attributes. The algorithm.
pre-college dataset features examined in this study is grouped During the feature selection process, no specific form of
into two: demographic and socio-economic (gender, blood relationship is assumed. The outcome of the feature selection is
type, skills, sports, musical instrument, province of origin, list of predictors ranked according to their importance.
parents educational background, parent‟s income, parent‟s
tribe, religion, number of brothers, number of sisters and rank 1) Information Gain Ratio (IGR): The first FSM
in family) and academic potential (admission test score in employed is the IGR. It calculates the entropies in class and
Math, Language Usage, Aptitude, and Science). On the other resolves the vulnerability of IG. Fig. 4 shows the code snippet
hand, post-admission data are educational achievement of the method used in this study.
indicators such as course, scholarship status, grades in Math 2) Correlation feature selection: CFS finds attribute that
and English subjects, and grade point average of first semester. are highly related with the specific groups but still have at
least inter-correlation amongst the attributes themselves. Fig.
5 shows the code snippet of the method used in this study.
271 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
D. Data Modelling
A prediction model for EDM can be developed using EDM
techniques but will heavily depend on the type of datasets. In
this study, logistic regression method is used.
The dataset is partitioned into training and validation
subsets. Two predictive models were created, for the first
Fig. 4. IGR Code Snippet.
model all the features will be inputted. On the second model,
only the significant variables assessed by feature selection
techniques were the final parameters in creating the model.
70% of the dataset is used in training and the remaining 30% is
used as a test-set for both models and are tested for accuracy
using 10-fold cross-validation.
E. Result Evaluation
The performance of the two models is evaluated by its
accuracy and precision which are computed using the equation
below.
(5)
(6)
Fig. 5. CFS Code Snippet.
Precision is computed by the positive predicted instances
over the total predicted instances.
IV. RESULTS AND DISCUSSIONS
A. Results of Feature Selection
Fig. 7 shows the result of IGR. The result is based on the
upmost Gain ratio ranked by their importance. Any information
gain above zero shows some type of significance. Factors like
English status, Math status, family income and college
entrance score for language usage, math, aptitude and science
largely influence the result of student‟s retention.
Fig. 8 shows the features and Correlation-based Feature
Selection scores ranked in ascending order of importance.
272 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Among the highest ranked by CFS are English status, gross TABLE. I. RANKED OF PREDICTOR VARIABLES IN ASCENDING ORDER
income and math status.
CFS Info Gain Ratio ChiSquared
Fig. 9 results show features that were highly influential or firstsemGPA firstsemGPA firstsemGPA
with high chi-square values. These values are displayed in
ENGLISHSTATUS ENGLISHSTATUS ENGLISHSTATUS
ascending order.
MATHSTATUS MATHSTATUS MATHSTATUS
PAYING BROTHERSNUM PAYING
LU GrossIncome LU
RELIGIONCATEGOR RELIGIONCATEGOR
APTITUDE
Y Y
SAME PROVINCE LU SAME PROVINCE
SCIENCE MATH APTITUDE
APTITUDE SAME PROVINCE SCIENCE
BROTHERSNUM SCIENCE MATH
MATH SISTERSNUM GrossIncome
RELIGIONCATEGOR
SISTERSNUM CanPlay
Y
Fig. 7. Attributes Ranked by Information Gain Ratio.
GrossIncome PAYING SISTERSNUM
CanPlay CanPlay BROTHERSNUM
273 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In future, the study can be enhanced by applying few [9] U. Shafique, and H. Qaiser, “A Comparative Study of Data Mining
hybrid feature selection algorithms on student datasets in order Process Models (KDD, CRISP-DM and SEMMA),” International
Journal of Innovation and Scientific Research, Vol. 12 No. 1 Nov. 2014.
to predict student retention. A web-based system will be
[10] P. Ramya, K. Gudlavalleru and M. Kumar, “Student Performance
developed that helps to monitor students and accurately predict Analysis Using Educational Data Mining,” International Journal of
student retention and attrition. Computer Science and Information Security (IJCSIS), Vol. 14, 2016.
[11] J. Pattanaphanchai, K. Leelerpanyakul, & N. Theppalak, ”The
ACKNOWLEDGMENT Investigation of Student Dropout Prediction Model in Thai Higher
The author would like to thank the Commission on Higher Education Using Educational Data Mining: A Case Study of Faculty of
Science, Prince of Songkla University,” Journal of University of
Education (CHED) for supporting this study under the Babylon for Pure and Applied Sciences, Vol.(27), No.(1): 2019.
Individual Research Grants (CHED M.O. No. 14-2016). [12] F. Ahmad, N. Ismail, and A. Aziz, “The Prediction of Students‟
REFERENCES Academic Performance Using Classification Data Mining Techniques,”
[1] National Center for Education Statistics, “The Condition of Education Applied Mathematical Sciences, vol. 9, no. 129, pp. 6415-6426, 2015.
2016.” (NCES 2016-144), Undergraduate Retention and Graduation [13] W. Bhaya, “Review of Data Preprocessing Techniques in Datamining,”
Rates, 2016. Journal of Engineering and Applied Sciences, 12 (16): 4102-4107, 2017.
[2] A. Norton, and I. Cherastidtham, “Dropping out: the benefits and cost [14] A. Algarni, “Data Mining in Education,” (IJACSA) International Journal
of trying university”, Grattan Institute, 2018. of Advanced Computer Science and Applications, Vol. 7, No. 6, 2016.
[3] Commission on Higher Education (CHED), “2018 Higher Education [15] M. Paraiso, H. Torres, et al., “Data Mining Approach for Analyzing
Facts and Figures,” 2018. Graduating Students‟ Academic Performance of New Era University –
[4] Philippine News Agency, “Only 23% of Filipinos finish college,” Bachelor Science in Computer Science”. International Journal of
BusinessMirror, (April 27, 2017). Conceptions on Computing and Information Technology. Vol. 3. Issue
3, 2015.
[5] Australian Government Department of Education and Training,
“Improving retention completion and success in higher education,” [16] M. Hall, and L. Smith, “Feature Selection for Machine Learning:
Higher Education Standards Panel Discussion Paper, June 2017. Comparing a Correlation-Based Filter Approach to the Wrapper,”
Proceedings of the Twelfth International Florida Artificial Intelligence
[6] N. Garett, M. Bridgewater and B. Feinstein, “How Student Performance Research Society Conference, 1999.
in First-Year Composition Predicts Retention and Overall Student
Success,” Retention, Persistence, and Writing Programs, Louisville, CO: [17] Sheena, K. Kumar, and G. Kumar, “Analysis of Feature Selection
University Press of Colorado, 2017. Techniques: A Data Mining Approach,” International Journal of
Computer Applications, 2016.
[7] P. Van der Zanden, E. Denessen, A. Cillesen and P. Meijer, “Domains
and predictors of first-year student success: A systematic review,” [18] J. Novaković, P. Strbac, and D. Bulatović, “Toward optimal feature
Educational Research Review, 23 57-77, 2018. selection using ranking methods and classification algorithms,”
Yugoslav Journal of Operations Research 21, 2011.
[8] A. Seidman, “College student retention: formula for student success,”
Westport, CT: ACE/Praeger, 2005. [19] M. Trabelsi, N. Meddouri, and M. Maddouri, “A New Feature Selection
Method for Nominal Classifier based on Formal Concept Analysis,”
Procedia Computer Science, 2017.
274 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This technical quest aspired to build deep should be put in place. Otherwise, it would remain unmitigated
multifaceted system proficient in forecasting banana harvest or even intensified and could suffer shortfalls and will continue
yields essential for extensive planning for a sustainable to exert pressure on agricultural produce.
production in the agriculture sector. Recently, deep-learning
(DL) approach has been used as a new alternative model in Yields from crops play a noteworthy role in the economic
forecasting. In this paper, the enhanced DL approach progression. Among the major fruit crops, banana (Musa sp.) is
incorporates multiple long short term memory (LSTM) layers one of the important tropical fruit crops and part of the rising
employed with multiple neurons in each layer, fully trained and economy of many developing countries like Philippines and
built a state for forecasting. The enhanced model used the the world’s most important goods following rice, wheat and
banana harvest yield data from agrarian reform beneficiary maize. Hence, yield assessment of banana production is
(ARB) cooperative of Dapco in Davao del Norte, Philippines. The essential for policy decisions regarding procurement,
model parameters such as epoch, batch size and neurons distribution, buffer stocking, import-export, price-fixation and
underwent tuning to identify its optimal values to be used in the marketing [3]. In view of that, more accurate forecasts of the
experiments. Additionally, the root-mean-squared error (RMSE) harvest yields and crop production provides an aid to an
is used to evaluate the performance of the model. Using the same effective and efficient decision making using timely
set of training and testing data, experiment exhibits that the information. It is a significant phase for an emerging economy
enhanced model achieved the optimal result of 34.805 in terms of so that adequate planning is undertaken for sustainable growth
RMSE. This means that the enhanced model outperforms the
[4] and for the overall development of the country. However,
single and multiple LSTM layer with 43.5 percent and 44.95
percent reduction in error rates, respectively. Since there is no
studies have shown that agricultural problems like forecasting
proof that LSTM recurrent neutral network has been used with yields remain difficult due to the lack of the necessary
the same agricultural problem domain, therefore, there is no infrastructures and there is no proof of optimal model to handle
standard available with regards to the level of error reduction in time series (TS) data to be used in forecasting such as the
the forecast. Moreover, investigating the performance of the banana harvest yields dataset.
model using diverse datasets specifically with multiple input Previously, different conventional models such as
features (multivariate) is suggested for exploration. Furthermore,
autoregressive integrated moving average (ARIMA) [5], [6],
extending and embedding this approach to a web-based along
with a handy application is the future plan for the benefit of the
[7], [8] are used in forecasting more specifically using TS data.
medium scale banana growers of the region for efficient and However, a major drawback in its used in forecasting is its
effective decision making and advance planning. incapability to recognized nonlinearities [9], [10], [4]. These
classical techniques have been replaced by DL algorithms [11].
Keywords—Yield forecasting; Deep Learning; Long short-term DL approaches are capable of identifying non-linearity and
memory; Banana harvest yield forecasting complexity of data [12], [8] in TS forecasting. Hence, this
advance approach is referring to as a future promising tool [13]
I. INTRODUCTION in forecasting yields in the field of agriculture.
Deep learning (DL) is a method that has been enticing The more recent structures of DL are Deep Neural
attention in recent years of machine learning and its continuous Networks, Convolutional Neural Networks, Recurrent Neural
growth gains more popular among researchers in diverse Networks (RNN), and Q-learning [11]. Among these DL
disciplines [1] where advancement and progression are fast and architecture, RNN model presents elevated performance in
incremental. Frequently, development takes place in a well- prediction, as it can capture the time features and the
resourced area (e.g. medical, security) and this budding architectures demonstrate dynamic temporal behavior [14].
application of DL is dispersed to the agricultural sector [2]. In However, the training is difficult and the major weakness of
agriculture, production is an essential phenomenon in natural RNN is carried out during the requirement of learning long-
aspect where progressions fuse in multifaceted ways and range time dependencies [15]. This limitation is addressed
production patterns are specifically influence by market factors through the development of LSTM algorithm [16], [17], [18].
and relentlessly affected by extreme events (i.e. floods, LSTM is developed to seize the “vanishing gradient problem”
droughts) and revealed to deteriorate and decreases its yield. encountered in RNN. It is also capable of learning long term
Thus, management technologies and timely interventions dependences [19], [10]. It is the state-of-the-art technique for
275 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
sequence learning [20], [21], [16] and TS prediction such as in B. Time Series Forecasting Applications, Challenges and
financial market [22], hydrology [19], petroleum production Methods
[23], energy [24], [25], [26], [27], neurocomputing [28], [24], Investigating time series and dynamic demonstration is an
[18], [17], expert systems [16], internet of things [10]. interesting exploration. Analysis of TS data aims to research
However, the weakness of this technique has perceived as it the observations trail and construct a model to depict the
did not perform satisfactorily in dealing with TS forecasting. structure of data and forecast future values. Hence, it is vital to
Its shallow architecture makes it incapable to exemplify the conceptualize an applicable model aiming at improving
complex characteristics of TS data more specifically in accuracy of the forecast. There are some different domains
handling extremely nonlinear and long interval TS datasets which already tested the capacity and adopted the used of
[15] such as in banana harvest yields data. Furthermore, this LSTM in TS forecasting problems such as predicting
limitation compels LSTM to be unclear if it is the best design emergency event occurrence [16] which solves classification
to work out real problem especially in using the harvest yield and regression problem and exhibits better performance which
dataset and the optimization issues due to the size of the data is proven effective over conventional methods. In forecasting
and the model tuning strategy applied. Also, LSTM has not petroleum production, [15] which case study involves the
been used in forecasting harvest yields such as in bananas since production of two separate oil depot at a particular time period,
its inception. proves the capability and eligibility of LSTM to be applied in
In this paper, the enhanced deep learning-based approach is the nonlinear forecasting problems and outperforms outmoded
used in forecasting harvest yields of banana production. The and traditional mathematical forecasting models. In predicting
performance of the model is then evaluated in terms of traffic flow [24], where it involves short term traffic flow at a
accuracy measures. The result of this study will be a great time interval of 1 to 5 minutes, LSTM and Attention
contribution to the consistent management for the improvement Mechanism shows excellent performance in dealing with 5- or
of harvest yields and to the overall production. It would also 1-minute-long historical data. However, the performance of the
provide a new technique to assist the agrarian reform model declines swiftly as the length of the sequence increases
beneficiary (ARB) cooperative of Dapco in its individual because of the collection of errors: the longer the sequence in
farming scheme, decision-making process and advance LSTM, the greater the error. Thus, it considers time and space
planning. features combination. Otherwise, in predicting the remaining
useful life of proton exchange membrane fuel cell [18], LSTM
The following parts of the paper are structured as follows: model quickly and accurately predicts the remaining service
Section II looks into the idea of Deep Learning, time series life and suitable for online residual life prediction but the
forecasting, application, and challenges. Section III features the robustness and generalizations performance need to be further
fundamentals of LSTM and the enhanced model. Section IV strengthened and improved. It also outperforms the outmoded
highlights the experiments as well as the outcomes and finally, techniques in predicting hourly day-ahead solar irradiance [25]
suggestions and conclusions are offered in Section V. though error is encountered when using weather forecast but it
shows less overfitting and better generalization. Thus, further
II. RELATED LITERATURE
evaluation and assessment is needed.
A. Deep Learning Approach
Moreover, it also shows simplicity and effectivity than
DL establishes a current, modern technique for data ARIMA and back propagation neural network in the tourism
analysis, with likely results and significant capability [29]. As flow prediction [10]. It suggests further that more hidden states
DL has been effectively applied in several domains, it has are to be tested and superiority over the classic feed forward
recently entered also in agriculture. Moreover, the performance neural network and the double-LSTM models in predicting
of the model is generally high and its potentials can also be water table depth [19]. A dropout strategy is being
applied in a wide variety of agricultural related problems not implemented to avoid overfitting though it has a resilience to
only involving images but also in forecasting TS data. The learn TS data but has insufficiency in its fitting ability. In
efficiency of its testing time is relatively faster than any other financial market predictions, it is effective in extracting
outmoded methods and the likelihood to develop simulated meaningful information from a noisy financial TS [22]
datasets to train model [13] is another advantage to solve real- compared to random forest, standard deep nets and logistic
world problems. regression and it turns out to be an advancement of the domain
DL algorithm consists of various components like RNN with respect to prediction accuracy and daily returns after
[11] and is the most popular method in forecasting task based transaction cost. It also outperforms conventional techniques in
on the solution intricacy, the desired accuracy in prediction and short-term load forecasts [26] and exhibits consistency for the
features of data [15]. More approaches are adopting RNN ambiguity with snowball in hours to forecast. LSTM proves
particularly LSTM exploiting the time dimension to perform appropriateness in TS modelling and forecasting with reduction
higher performance prediction [13] and more sophisticated in percentage errors and shows simplicity towards managing
architecture in dealing with large datasets which could improve the information instead of working over complex equations.
its performance [30], [31].
276 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
III. THE ENHANCED APPROACH and a pointwise multiplication and addition operation. These
gates and the memory cell allow an LSTM unit to respectively
This segment depicts the enhanced deep learning approach forget, memorize and expose [14] the memory content.
used in forecasting banana harvest yields from data preparation
to evaluating the forecast. Also, important steps are discussed It has an input at time step t denoted by (xt), and the hidden
and described in this section. state from the previous time step (St−1) that is introduced to
LSTM block, and then the hidden state (St) is computed
A. Data through the forget gate (ft), input gate (it) and output gate (ot)
The banana harvest yield data set features the number of where the input and forget gates are responsible of how much
stems cut, the number of boxes produce and box-stem ratio. new content should be remembered (memorized) and how
The author chose the “number of stems cut” as the harvest much old content should be disregarded (forgotten). Gates are
yields time series data to be used during the experiment. In this computed using the following set of formulas arrange in steps:
study, the term “stems” is referring to the whole bunch of the
banana fruit cut from the plant. The harvest data came from the Step 1: Determine what content is going to be disregarded
ARB cooperative of Dapco in Davao del Norte, Philippines. from the cell state which will be decided by the forget gate (ft):
The cooperative is one of the key players from the small- ft=sigmoid(Wxfxt+Wsfht-1+bf) (1)
medium scale banana growers in the region and has contributed
to the overall production on the exportation of bananas to other Step 2: Determine which new content is going to be kept in
countries like Japan. The dataset contains thirty-five thousand the cell state which will be decided in two phases: First, the
series of observations approximately from year 2014 to 2018 input gate(it) layer decides which values to be updated. Second,
where each year is composed of thirteen (13) periods and it a tanh layer that forms a vector of new candidate values (Ĉt).
usually starts from second half of the first month of the year to These two phases can be illustrated as follows:
the first half of the following year. it=sigmoid(Wxixt+Wsiht-1+bi) (2)
B. Data Preprocessing Ĉt = tanh (Wcixt+Wciht-1+bc) (3)
Series of data transformations are done before fitting the
model to the dataset and making a forecast. It includes Step 3: Update the previous cell state (Ct−1) into the new
converting the TS data into supervised learning to make it cell state (Ĉt), which can be conveyed as:
stationary. A lag differencing is used to strip off the increasing Ct = ft *Ct-1+it * Ĉt (4)
trend in the data. Transformation of data into a supervised
learning problem and scaling to values to meet the hyperbolic Step 4: Lastly, determine the desired output to be produced.
tangent activation function of the model is also done. All these The output will be a streamed form and will be based on the
transforms will be inverted back on forecast to revert data into cell state. The output gate (ot) in this step will decide what part
its original scale before evaluation and in determining the error of the cell state is going to be produced as output and then goes
score and splitting data into training and testing sets through the tanh layer, impelling values to be between -1 and
respectively. Aiming to obtain the best outcomes, iterative 1, and multiplying it to the output gate as illustrated in the
optimization is used which means attaining the outcomes equation below:
several times and select the utmost optimal iteration that has ot=sigmoid(Wxoxt+Whoht-1+bo) (5)
less errors. Each important parameter, such as epoch, batch size
and neurons, is given a varied value and the experiment for St=ot*tanh(Ct) (6)
each parameter are run in several repetitions as desired. The
best value for each parameter is identified through the The LSTM is represented with the two sets of parameters
summary of performance report using the RMSE scores from from the preceding six equations. These are: Wxf, Wsf, Wxi, Wsi,
each population of results. Wci, Wxo, Who which are referred to as weights and bf, bi, bc and
bo are biases, respectively.
C. The Multi-Dynamic Long Short Term Memory (mdLSTM)
Model
It is indispensable to briefly explain the fundamentals of
LSTM memory block as it is the precedent of the enhanced
model prior to its introduction.
1) Fundamentals of the Long Short Term Memory (LSTM)
Model.
LSTM was first proposed in 1997 (Sepp Hochreiter, Jürgen
Schmidhuber), driven by an analysis of error flow in prevailing
RNNs (Hochreiter et al., 2001) [16]. It shows suitability for
processing and forecasting using TS data. The LSTM block,
shown in Fig. 1, depicts a cell state (Ct) which resembles a
conveyor belt. It thoroughly takes turn in the chain and Fig. 1. The LSTM Memory Block, where is the Pointwise Addition,
controlled by constitution of gates; an elective inlet means of is the Pointwise Multiplication, Hyperbolic Tangent Activation Function
information. Gates are comprised of a sigmoid neural net layer and is the Sigmoid Activation.
277 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
2) The multi-dynamic Long Short-Term Memory IV. EXPERIMENTAL RESULTS AND DISCUSSION
(mdLSTM) The enhanced deep learning model was implemented using
An effective means to a better overall performance of the the banana harvest yields dataset. Given the nature of the
neural network is to augment its deepness [11]. The dataset, long interval time and the missing (zero) observations
development of expound LSTM recurrent neural network is are seen and considered. In view of the fact that missing values
encouraged by the remarkable learning capabilities of profound were observed, simply removing records containing zero value
recurrent network design to be utilized in TS forecasting is the simplest strategy ever devised during the experiment
applications. The enhanced model, shown in Fig. 2, has several with the goal of obtaining the optimal outcome of the enhanced
LSTM layers, heaped one after the other joined to blend the model.
advantage of a sole LSTM layer in an expound recurrent For consistency and fair assessment, splitting of data to
network manner feed with multiple value of neurons. The aim 80% training and 20% for testing is done. Moreover, the model
of the enhanced model is to construct the characteristics in a uses the optimal outcome from the lowest mean RMSE of the
hierarchical design where the lower layer separates the input model parameter tuning for the values to be assigned for
data disparity factors and these demonstrations are merged at epochs, batch size and neurons. Noticeably, the number of
the upper layer. Such deep structure will simplify well owing epoch is very minimal because there is no proof that suggests
to a trimmed representation than a shallow design in case of the exact number of cycles (epochs) to train a model.
large or complex datasets [32], [33] such as in the banana Experiment wise, it is evident that setting the value of epoch
harvest yields. and neurons minimally produces a sensible model outcome in
D. Training and Forecasting terms of reduction of errors and forecasting accuracy most
likely if using large datasets. The size of the diverse dataset is
To execute the model, Keras library along with Theano and considerably important in setting the number of epochs
Tensorflow backend are properly installed and configured, because it shows different behavior to optimally train the
splitting of dataset into 80% training and 20% testing is done network.
respectively to capture the thirteen (13) periods of the year
2018 as test dataset. The number of epochs, batch size and During the experiment, the model uses the optimal value of
neurons are assigned a minimal value from the summary result the parameters such as the number of epochs, batch size and
of the performance done in the tuning step. The “mean squared neurons with respect to the result obtains from the model
error” loss function with “ADAM” optimization algorithm are parameter tuning. Also, it has been noticed that executing the
used in compiling the model. RMSE [34], [26], [22], [8] is model more than once with the same parameter values does not
used to evaluate the performance of the model to forecast. It guarantee better performance and at some instances, worsen
calculates the variance between actual value and the predicted the result. There are three sets of experiment done in
value and used to evaluate different models for a certain data forecasting using banana harvest yield dataset: first, using
and not amid datasets. The following formula is used for single LSTM layer, second, using multiple LSTM layers
computing the RMSE: assigned with same value of neurons in each layer and third,
using the enhanced deep learning model where multiple layers
RMSE = √ ∑ (7) feed with multiple value of neurons. The latter used the
precedent numbers of the optimal value of neuron obtained
This measure is calculated by associating the target values from the model parameter tuning. The results of experiments
for the time series and its consequent time series predictions, are exhibited in Table I, Table II and Table III, respectively.
For uniformity, all experiments were done with and without a
where n is the total number of observations, is the actual
dropout rate. The second and third experiment used two up to
value; whereas, is the predicted value. four LSTM layers only.
278 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
279 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
280 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Dengue is a viral mosquito-borne infection that is had high dengue transmission rates resulting to more steady
endemic and has become a major public health concern in the water as potential breeding grounds for mosquitoes [3].
Philippines. Cases of dengue in the country have been recorded Humidity had been a consistent, significant weather factor that
to be increasing, however, it is reported that the country lacks provided favorable conditions for Dengue vectors [5].
predictive system that could aid in the formulation of an effective
approach to combat the rise of dengue cases. Various studies Despite the effort of the government to look for possible
have reported that climatic factors can influence the transmission ways to control the increase of dengue cases in the Philippines,
rate of dengue. Thus, this study aimed to predict the probability there is still no specific solution or response on how to control
of dengue incidence in Iligan City per barangay based on the the dengue outbreaks up to this writing. This circumstance
relationship of climatic factors and dengue cases using different necessitates the implementation of primary safety measures to
predictive models with data from 2008 to 2017. Multiple Linear reduce and prevent dengue infections, to control mosquito
Regression, Poisson Regression, and Random Forest are populations, and limit the spread of dengue cases nationwide.
integrated in a mini-system to automate the display of the Since climate conditions influence the dengue transmission
prediction result. Results indicate that Random Forest works cycle [6] [7] [8], a dengue risk-prediction system based on the
better with 73.0% accuracy result and 33.58% error percentage, relationship of dengue incidence and climatic conditions is
with time period and mean temperature as predictive variables. investigated in this study. The development of a risk prediction
system could forecast the locale of possible high incidence rate
Keywords—Dengue; predictive models; Pearson’s correlation;
multiple linear regression; Poisson regression; random forest
of dengue thus will have significant contribution in controlling
the spread of dengue by reducing the transmission of
I. INTRODUCTION mosquitoes [9].
Dengue has been an endemic infection in over 100 Predictive analytical approach using a variety of machine
countries in the world, in tropical and subtropical regions. One learning, modeling, statistics, artificial intelligence, and data
of the four dengue viruses had been classified as dengue mining algorithms could be input data to predict unknown
serotypes (DENV-1, DENV-2, DENV-3, and DENV-4), events in the future. Also, the use of statistical methods,
carried by the main vectors, the Aedes Aegypti and Aedes correlations between dengue incidence and climatic variables
Albopictus [1]. Reports were gathered from the World Health were established to predict potential outbreaks in specific
Organization (2018) that the Dengue fever was the most areas. Promprou [6] sampled a predictive model to predict the
critical and rapidly spreading mosquito-borne viral disease in Dengue Haemorrhagic Fever (DHF) in Thailand using
the world for over the past 50 years with 390 million dengue Multiple Linear Regression model to explain the relationship
infections per year in 3.9 billion people in 128 countries at risk between the household’s activities and DHF patients. Results
of infection. of the study revealed a 26.9% of the variation of DHF patients
using a number of water storage containers, Aedes Aegypti in
In the Philippines, Dengue fever has become one of the drainage of refrigerators, pH and temperature of water in
major health problems among the populace. The Department of container. In the study by Ong et al. [7] predicting the Dengue
Health (DOH) reported a total of 138,444 dengue cases incidence with the use of Random Forest approach, predicted
nationwide from January 1 to October 6 2018 representing the risk rank of dengue transmission in Singapore with dengue
21% increase on the number of cases in the same period in cases, population, entomological and environmental data. The
2017 [2]. Just recently, there were 115, 986 dengue cases evaluation using the latest dengue case data in the study
recorded in the Philippines including 491 deaths reported from showed a strong predictive ability for the model, compared to
January 1 to July 6 2019; 86 % higher than in 2018 [3]. Same the study results of Tilwani, Dave, & Nadurbarkar [8], that
report cited Region X as one of the regions with highest adopted a regression approach with Poisson Regression and
incidence rate having 9,354 cases [3]. Incidence of Dengue was Negative Binomial to investigate the correlation between
caused by several factors, one of which was the climatic dengue incidence and climatic fluctuations including relative
conditions referring to temperature, relative rainfall, and humidity, temperature, and pressure. Dengue cases with 70%
relative humidity were reported to be important influential accuracy showed the impact of climatic fluctuations in dengue
Dengue transmitters [4]. Studies discovered that places with transmission.
high temperatures and higher rainfall such as in the Philippines
281 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
282 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
D. Development of the Predictive Model were originally modelled for z, have to be transformed back
Correlation analysis was used to determine the strength of properly to a model for y for interpretation. To estimate the b
the relationships between the monthly number of dengue cases, values, a process called maximum likelihood estimation (MLE)
dependent variables and climatic factors (minimum or weighted least squares may be used where it fitted a model.
temperature (tmin), maximum temperature (tmax), mean Random forest is an ensemble of simple tree predictors
temperature (tmean), relative rainfall (rr), minimum relative used to determine the final outcome. It used a bootstrap
humidity (rhmin), maximum relative humidity (rhmax), and sampling approach to generate k different training data subsets
average relative humidity (rhmean)) and time period from an original dataset, and then k decision trees were
(timeperiod) as independent variables, covering 120 months in constructed by training these subsets. The final value was
total from 2008 to 2017. calculated by taking the average of all the predicted values by
To see how the data sets were correlated, Pearson’s all the trees in forest. Since there was no given regression
Correlation Coefficient was used. Given that the data are coefficients in building the Random Forest model, variable
continuous, this method is suitable to perform as it is generally selection was done through performing the Feature Importance
used when variables are continuous in nature such as in ratio or method, which variables had the most effect on the model.
interval scale variables. Pearson’s correlation coefficient is Visualizing tree in Random Forest was to evaluate the
indicated by r and defined by: predictions for each row using all the trees in the model, how
each variable contributed to the final prediction. To calculate
(1) for the result, propents have used the RandomForestRegressor
√* ( ) +* ( ) +
class of the sklearn.ensemble library in Python.
The value of r always ranges from -1 to +1. The However, before performing the prediction process in
relationship between the variables is said to be not related MLR, selecting highly significant independent variables based
when the value of r comes down to 0. If the value of r lies to on their p-value was done first in order to build a best model,
+1, then the variables are said to be positively correlated, while the called the Stepwise Regression.
variables are said to be negatively correlated if the value of r is -1.
Dengue Prediction System: The proposed system was
The development of the forecasting model is based on
implemented using Python and PostgreSQL as the backend,
Multiple Linear Regression (MLR), Poisson Regression, and
while the front end was developed using CSS, Javascript,
Random Forest.
HTML, Leafleat, and Django. This system will automate the
Multiple Linear Regression analysis is a statistical prediction of the model chosen, provided that data are
technique that used several explanatory (independent) variables presented in the correct format.
to predict the outcome of a response (dependent) variable. The
model is developed with the following equation. IV. RESULTS AND DISCUSSION
As shown in Fig. 3, most of the values were not close to 1
y = b0 + b1x1 + b2x2 + b3x3 + ... + bnxn (2)
which indicates that there is a weak correlation between
Where independent variables. On the other hand, timeperiod
associated with cases has an r of 0.3318 and has the greatest
bi = y-intercept correlation coefficient among the other independent variables.
y = dengue cases Other independent variables such as tmin and tmean have
positive correlation with dengue cases having r values equal to
xi = climatic factors and time period 0.1439 and 0.1958, respectively while there is negative
correlation with rhmin and rhmean with cases havingr equal to
The value of R-squared and adjusted R-squared were
-0.1298 and -0.1424 consecutively.
calculated in order to test how well the data fit the regression
model. Usually, the higher R-squared value between 0 and 100 Result of the Multiple Linear Regression displays in Fig. 4,
means the better the regression model fitted the observations. shows that r2 is equal to 0.2304. This indicates that all eight of
To identify the significance of each of the independent the independent variables explain 23.04% of the variability on
variables, P-value was also computed. the number of dengue cases. Moreover, it was shown that some
independent variables had a p-value greater than 0.05 lowering
The Poisson Regression on the other hand is designed to fit
their significance in predicting dengue incidence. From this
a model of regression in which counts were made with the
model, only the maximum temperature, average relative
dependent variable Y (dengue cases). The fitted model Y to
humidity, and time period have p-values less than 0.05; thus,
one or more X predictor variables (climatic factors and time
have significant impact on dengue cases.
period), which were either quantitative or categorical. A
poisson regression model defined as: Stepwise regression, as seen in Fig. 5, was conducted to
eliminate non-significant variables to build a reliable model out
z = e + b0 + b1x1 + b2x2 + b3x3 + ... + bkxk (3)
of the remaining independent variables. The variables tmin,
where z = log(y) , transformed from y in generalized linear rhmax, tmean, and rhmin were removed from the model. These
modelling called link function. This was done so that a linear variables were omitted in building the Multiple Linear
regression modelling in z satisfied all required assumptions Regression since their respective p-values were higher than
[11]. The corresponding coefficients (b1, b2 … bk), which 0.05.
283 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
284 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
predicting the number of dengue incidence based on climatic prediction = regressor.predict (X_test)
factors. from sklearn.metric import r2_score
print(‘R-Squared score is: {θ}%’.format(round(r2_score(y_test,prediction) * 100, 2)))
Fig. 7. The Actual Values and Predicted Number of Dengue Incidence from
Time Period of January 2008 to December 2017 for Multiple Linear
Regression.
Multiple Linear
18% 67.14%
Regression
Fig. 8. Feature Importance Performed for Random Forest. Fig. 11. A Screenshot of the Developed Dengue Prediction System.
285 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
286 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Network on Chip (NoC) is an emerging operating in any of the Open system interconnection (OSI)
interconnect solution with reliable and scalable features over the layers. The multifunctional intercommunication supported by
System on Chip (SoC) and helps to overcome the drawbacks of the Gateway on a single-core chip. The different protocol
bus-based interconnection in SoC. The multiple cores or other standards like Bluetooth, Modbus, serial bus, Process Filed
networks have a boundary which is limited to communicate with bus, and Controller area Network (CAN) provides
devices, which are directly connected to it. To communicate with intercommunication using gateway [4]. The intelligent
these multiple cores outside the boundary, the NOC requires the Gateway has interoperation and achieves better communication
gateway functionality. In this manuscript, a cost-effective among different bus networks with reconfigurability and also
Network Gateway (NG) model is designed, and also the
supports fast conversion speed, flexibility, intellectual control
interconnection of a network gateway with multiple cores are
ability, reliability, and higher-level interface. The protocol
connected to the NoC based system is prototyped on Artix-7
FPGA. The NG mainly consists of Serializer and deserializer for converting Gateway works on Most of the OSI layers [5-6].
transmitting and receiving the data packets with proper The high-performance computation needs high-speed
synchronization, temporary register to hold the network data, interconnection like Ethernet and Infiniband. The data
electronic crossbar switch is connected with multiple cores which transmission between two heterogeneous networks needs an
are controlled by switch controller. The NG with the Router and efficient network gateway to improve system performance in
different sizes of NoC based system is designed using congestion- terms of bandwidth and Latency [7] [12].
free adaptive-XY routing. The implementation results and
The gateway terminology is used commonly for most of the
performance evaluation are analyzed for NG based NoC in terms
of average Latency and maximum Throughput for different
applications for protocol conversion and data packets transfers.
Packet Injection Ratio (PIR). The proposed Network gateway The network gateways are used in most of the real-time
achieves low latency and high throughput in NoC based systems embedded and Internet of Things (IoT) applications. The home
for different PIR. gateway requires a standard ARM chip with SoC chip which
integrates the Customer Electronics Bus (CEbus) with home
Keywords—Network gateway; network on chip; FPGA; appliances like TV, microwave oven, refrigerator, and washing
routing; network interface; crossbar switch machine. The user sends a command to the internet; the
network control module receives the command, issues request
I. INTRODUCTION signal to Chip to control the home appliances [8]. The
The NoC will play an emerging role in future high- heterogeneous Gateway provides different interfaces to
performance Chip Multi-Processor (CMP) to address the internet, GSM, CDMS, PSTN, and so on, to support different
problems of interconnections. In recent years, most of the application scenarios [9]. The embedded Gateway is a
research focused on a packet-switched NoC design, which backbone for smart grid home networks [10], wireless
improves the system performance by using optimization applications [11] [14], indoor high precision positioning
techniques in the network for better Latency and bandwidth systems [13], and IoT applications [15] for communicating
and also supports on-chip and off-chip communications. The with other networks.
NoC based photonic communications support a mechanism for In this manuscript, a cost-effective Network Gateway
large data transmission with higher bandwidth and less power model is designed along with Gateway based NoC system
consumption. The photonic based NoC supports Multiple cores using Adaptive XY Routing. The Network gateway results are
interface using gateway switch [1] [3]. Multiple cores residing hardware resource-efficient, works at low Latency, and High
in a single chip (MPSoC) exist towards mixed-criticality Throughput for input traffics which are evaluated for NoC
system includes dependability, security, and different block based system. Section II explains about related work on
access with shared resources. The outside network real-time Gateway mechanism used for different applications and also
messages are communicating to MPSoC using a gateway [2]. explains about research findings. Section III elaborates the
In general, the network gateway is a node which connects two Network Gateway architecture using electronic crossbar switch
different networks with different transmission protocols and with an explanation. Section IV explains the Network Gateway
simplifies the internet connectivity into one electronic device. based NoC based system with router architecture. The results
The gateway node acts as a firewall and proxy server for and performance evaluation are analyzed with tables and
business use. Gateways are a protocol which provides the graphs in Section V. Finally concludes the overall proposed
compatibility between two different protocols and will be work with Future scope in Section VI.
287 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
II. RELATED WORKS system (CPPS) gateway is described by Urbina et al. [26] to
In this section, the general Gateway related work and meet the industry 4.0 standards. The industrial network
applications of Gateway are reviewed. Shi et al. [16] presented architecture includes CPPS Gateways, which are
an embedded dual home architecture with secured Gateway interconnected with multiple peripherals, electronic and
both on hardware and software platform. The Gateway electrical devices using different network protocols like
improves the transmitting information risk by the user and Profinet, Profibus, and High availability Seamless Redundancy
network isolation module using FPGA is incorporated to (HSR). Kwak et al. [27] present the trust domain gateway
improve the security features using data signature and key system to solve the untrusted internet structural problems.
management. The secured embedded Virtual private network Gaps in the research: Most of the work carried on
(VPN) gateway is presented by Han et al. [17] to improve the traditional software-based gateway designs lacks with latency
data transmission security with protection capability in and throughput issues. Hardware-based Network gateway
application terminals. This VPN gateway is worked under L3, designs use bus-based interconnections for embedded real-time
L4, and L7 layers with firewall protection, VPN Functioning, applications and lack of scalability and reliability problems.
and network isolation modules. Ajami et al. [18] presented an The existing research work is done on protocol conversion
FPGA Based embedded network firewall which supports using gateways, but not on NoC based system. In order to
highly customized data packet filtering on a network gateway. resolve these problems, a cost-effective Network gateway with
These firewall customized in real-time by changing the NoC based system is designed.
TCP/UDP port id, Source MAC address, and source-
destination IP address. Abuteir et al. [19] introduced a gateway III. NETWORK GATEWAY DESIGN
design to establish the hierarchical platform for multi-core The Gateway provides the network and access information
chips interaction either on on-chip or off-chip networks. The to the four gateway cores, and the hardware architecture of the
software-based Gateway supports message classification, network gateway is represented in Fig. 1. The network gateway
message – scheduling, traffic shaping services, downsampling, mainly consists of deserializer and Serializer for receiving and
service, protocol conversion, egress-queuing, ingress-Queuing, transmitting the data information’s with proper
Virtual-Link queuing, and also supports serialization services. synchronization, Temporary register, Electronic crossbar
Obermaisser et al. [20] described the mixed-criticality switch, Switch controller, priority encoder, and four gateway
systems for end to end real-time communication, which cores. The gateway cores are processors, buffers, caches,
involves gateways between multiple off-chip networks, peripheral devices, etc. The FIFO buffers are considered in the
Gateway between off-chip and on-chip networks. The gateway design.
node resolving the contention between source controlled and The data information is received from the network either
autonomous networks and also supports end-to-end addressing from the interface or from the Router to deserializer, which
and routing. The cloud storage gateway was presented by receives the data signals serially, works based on Serial In
Dumitru et al. [21] on FPGA platform. The secured data Parallel Out (SIPO) manner. The received 8-bit data converts
encryption and transparency are resolved by using FPGA to 32-bit data to parallel using shifting operation along with
between host and outside interface in cloud infrastructure. Lee issuing the synchronization signal to Serializer. The
et al. [22] presented a high-performance hardware-software synchronization is achieved between Serializer and deserializer
based gateway design for In-Vehicle Network (IVN) for CAN/ using counter method and proper clocking mechanism. The
FlexRay controllers. The data conversion between CAN to temporary register receives the deserialized data, holds for
FlexRay and vice-versa is achieved using Routing table access to the electronic crossbar. This temporary register is
converter block with AXI interface on Zed board. Shreejith et only used to store the received deserialized data signals and
al. [23] described the vehicular Ethernet Gateway connected that are scheduled towards for the gateway cores through
with multiple network protocols like FlexRay, CAN, and switch controller. The electronic crossbar switch receives the
Ethernet with embedded computing Units. The Ethernet temporary data along with gateway core (buffer) inputs and
gateway is designed using Switch fabric between FlexRay and works based on switch controller, and its hardware architecture
Ethernet controller. The switch fabric is designed using is represented in Fig. 2.
Crossbar switch.
Temporary
The embedded Gateway for Fourth Generation (4G) mobile Deserializer
Register
network and process Fieldbus (PB) with decentralized
Periphery (DP) is described by Zhou et al. [24] on FPGA Priority
platform. The AES algorithm is used for secured data Networ Encoder
Synchroni Electronic
transaction in Gateway. The Gateway is used to connect two (Interfac
zation Crossbar switch
e/ Switch
different protocol 4G and PB conversion in terms of data. The Router)
Korona et al. [25] introduced an Internet Protocol security Controller
(IPsec) gateway for multi-gigabit networks which includes
security association database to store secure information, Buffer Buffer Buffer Buffer
Serializer
Internet key exchange to set secure channels, and responsible 1 2 3 4
for all security operation with packet encapsulation. The
Fig. 1. Hardware Architecture of Network Gateway.
programmable-SoC (PSoC) based cyber-physical production
288 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Core
Inputs Network
Input Processing Network
Core Router Local links
Gateway
G G G G
Switch
Controller
R1 R5 R9 R13
to support any of the64- processing core information’s that can Define Dest.
transmit to any of the 16 routers via network gateways using XY Addr
Li
Adaptive routing algorithm. Ei
Arbitration
& Priority
Wi Input Encoder
The network gateway with cores is connected to routers via Register
Adaptive-XY
a network interface (NI). In design, Mesh topology is selected Si Routing
to design 2x2, 3x3, and 4x4 NoC architectures. In Fig. 3, the Ni
Define Congestion
4x4 NoC has 16 routers (R1 to R16), and all the routers are Parameters
interconnected using linked wires. All the network Gateway Output Register
with cores inputs are received to the corresponding routers via
the network interface and perform the data transaction based on E
Wo So No Lo
o
the destination address of the corresponding routers.
Fig. 4. Hardware Architecture of Network Gateway based Router.
289 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
1-bit 2-bits 2-bits 8-bits respectively for 5000MHz clock frequency. The network
gateway based NoC designs are implemented effectively on
Request Dest.X Dest.Y Gateway Data FPGA with better chip area, speed, and power tradeoffs have
been achieved.
Fig. 5. Packet formation for Network Gateway based Router.
TABLE. I. NETWORK GATEWAY RESOURCE IMPLEMENTATION RESULTS
The framed packet, along with four more from input
Resources Utilized on Artix-7 FPGA
register is input to priority encoder. The priority encoder works
based on the arbitration. The Arbiter receives the MSB bits Slice Registers 450
from all the five ports and considered as requests and generates Slice LUTs 893
the 5-bit grants based on the priority. These grants are acts as a LUT-Flipflops 252
select line to priority encoder. The encoded data is a prioritize Max. Frequency (MHz.) 319.642
packet data, and it sends as an input to the adaptive routing-XY Total power (W) 0.104
algorithm. Each Router, R1 to R16, has fixed 4-bit current XY
address and which is easy to identify the Router. For example,
TABLE. II. RESOURCE UTILIZATION–FOR NG-NOC DESIGN
in design, R4 is set to “0011," and R14 is “1101”.
Network Gateway –NoC designs
To perform the routing computation, first, define the
Area Utilization NG NG-2X2 NG-3X3 NG-4X4
congestion parameters along with Destination-XY address
Router NoC NoC NoC
from the encoded packet. The adaptive–XY routing is
congestion-free routing and adaptive form of normal –XY Slice Registers 470 603 823 1193
routing [28]. The X or Y direction with less number of routing Slice LUTs 914 1011 1216 1551
path is defined and the routing packet id assisted to the LUT-FF pairs 273 365 561 891
destination with less congestion. Based on congestion
parameters, which finds the shortest routing path to reach the
destination with less traffic. The Network Gateway based NG-Router NG-NoC 2X2
single router, 2X2, 3X3, and 4X4 NoC’s are designed and NG-NoC 3x3 NG- NoC 4x4
prototyped on FPGA, which are explained detail in the next 1600
section.
1400
V. RESULTS AND PERFORMANCE ANALYSIS 1200
Resources Utilized
290 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
291 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[4] Hu, Yonghong, and Lu Ding. "Design and Realization of Multi- In 2016 International Conference on Robots & Intelligent System
functional Gateway Based on Single Chip." In 2009 2nd International (ICRIS), pp. 159-163. IEEE, 2016.
Congress on Image and Signal Processing, pp. 1-4. IEEE, 2009. [17] Kun Han, Junjie Liu, Demin Yang and Quan Yuan, "The design of
[5] Guo, Tiantian, Ming' a Zhou, and Qing Shen. "A Reconfigurable secure embedded VPN gateway," 2014 IEEE Workshop on Advanced
Intelligent Gateway for Heterogeneous Networks." In Proceedings of Research and Technology in Industry Applications (WARTIA), Ottawa,
2013 Chinese Intelligent Automation Conference, pp. 485-493. ON, 2014, pp. 350-353.
Springer, Berlin, Heidelberg, 2013. [18] Ajami, Raouf, and Anh Dinh. "Embedded network firewall on FPGA."
[6] Guo, Xiaodong, and Haijun Ren. "Multimode communication gateway In 2011 Eighth International Conference on Information Technology:
design in heterogeneous network environments, intelligent distribution, New Generations, pp. 1041-1043. IEEE, 2011.
and utilization." In 2015 6th IEEE International Conference on Software [19] Abuteir, Mohammed, Romn Obermaisser, Zaher Owda, and Thierry
Engineering and Service Science (ICSESS), pp. 326-329. IEEE, 2015. Moudouthe. "Off-chip/on-chip gateway architecture for mixed-criticality
[7] Shi, Wei, Gaofeng Lv, Zhigang Sun, and Zhenghu Gong. "HiTrans: An systems based on networked multi-core chips." In 2015 IEEE 18th
FPGA-Based Gateway Design and Implementation in HPC International Conference on Computational Science and Engineering,
Environments." In International Conference on Algorithms and pp. 120-128. IEEE, 2015.
Architectures for Parallel Processing, pp. 561-571. Springer, Cham, [20] Obermaisser, Roman, Zaher Owda, Mohammed Abuteir, Hamidreza
2015. Ahmadian, and Donatus Weber. "End-to-end real-time communication
[8] Ni, Binbin, Mingguang Wu, and Yanpeng Liu. "Design of Embedded in mixed-criticality systems based on networked multi-core chips." In
Home Network Gateway for CE Bus Based on ARM." In 2006 4th IEEE 2014 17th Euromicro Conference on Digital System Design, pp. 293-
International Conference on Industrial Informatics, pp. 1380-1384. 302. IEEE, 2014.
IEEE, 2006. [21] Dumitru, Laurențiu A., Sergiu Eftimie, and Dan Fostea. "An FPGA-
[9] Luo, Hong, Cheng Chang, and Yan Sun. "Advanced sensor gateway Based cloud storage gateway." In 2nd International Conference SEA-
based on FPGA for wireless multimedia sensor networks." In 2011 CONF, Academia Navală Mircea Cel Bătrân, Constanța. 2016.
International Conference on Electric Information and Control [22] Lee, Trong-Yen, Chia-Wei Kuo, and I-An Lin. "High performance
Engineering, pp. 1141-1146. IEEE, 2011. CAN/FlexRay gateway design for in-vehicle network." In 2017 IEEE
[10] Nguyen, Minh-Triet, Lap-Luat Nguyen, and Tuan-Duc Nguyen. "On the Conference on Dependable and Secure Computing, pp. 240-242. IEEE,
design of gateway node for smart gird home network." In 2015 2017.
International Conference on Communications, Management and [23] Shreejith, Shanker, Philipp Mundhenk, Andreas Ettner, Suhaib A.
Telecommunications (ComManTel), pp. 57-61. IEEE, 2015. Fahmy, Sebastian Steinhorst, Martin Lukasiewycz, and Samarjit
[11] Shaofeng, Lin, Tao Bo, Pan Jin, Wan Juan, and Du Jia. "Design and Chakraborty. "VEGa: A high performance vehicular Ethernet gateway
Implementation of Embedded Wireless Gateway." In 2015 International on hybrid FPGA." IEEE Transactions on Computers 66, no. 10 (2017):
Conference on Intelligent Transportation, Big Data and Smart City, pp. 1790-1803.
270-273. IEEE, 2015. [24] Zhou, Yuan, Wenping Xiao, Mingshan Liu, and Xiaokun Li. "Design of
[12] Zheng, Qi. "The design and the implementation of communication the embedded gateway for 4G and PROFIBUS-DP based on FPGA."
gateway between CAN bus and Ethernet." In 2015 IEEE Advanced In 2017 3rd IEEE International Conference on Computer and
Information Technology, Electronic and Automation Control Communications (ICCC), pp. 748-752. IEEE, 2017.
Conference (IAEAC), pp. 862-866. IEEE, 2015. [25] Korona, Mateusz, Krzysztof Skowron, Mateusz Trzepiński, and Mariusz
[13] Wang, Wenhua, Weiwei Xia, Rui Zhang, and Lianfeng Shen. "Design Rawski. "FPGA implementation of IPsec protocol suite for multigigabit
and implementation of gateway and server in an indoor high-precision networks." In 2017 International Conference on Systems, Signals and
positioning system." In 2014 IEEE 3rd Global Conference on Consumer Image Processing (IWSSIP), pp. 1-5. IEEE, 2017.
Electronics (GCCE), pp. 540-541. IEEE, 2014. [26] Urbina, Marcelo, Armando Astarloa, Jesús Lázaro, Unai Bidarte, Igor
[14] Baoxia, Sun, Wang Weixing, Tie Fenglian, and Weng Jiangpeng. Villalta, and Mikel Rodriguez. "Cyber-physical production system
"Design and implementation of gateway for hybrid antenna clustering gateway based on a programmable SoC platform." IEEE Access 5:
routing algorithm in paddy monitoring." In 2016 6th International 20408-20417, 2017.
Conference on Electronics Information and Emergency Communication [27] Kwak, Byeong Ok, and Tae Soo Chung. "Design and Implementation of
(ICEIEC), pp. 310-313. IEEE, 2016. the Trust Domain Gateway System." In 2018 International Conference
[15] Zhong, Chang-Le, Zhen Zhu, and Ren-Gen Huang. "Study on the IOT on Information and Communication Technology Convergence (ICTC),
architecture and gateway technology." In 2015 14th International pp. 925-927. IEEE, 2018.
Symposium on Distributed Computing and Applications for Business [28] Guruprasad, S. P., and B. S. Chandrasekar, "Design and Performance
Engineering and Science (DCABES), pp. 196-199. IEEE, 2015. analysis of Adaptive-XY over N-XY and DO Routing on FPGA
[16] Shi, Yonghong, Jianzhong Shen, Lin Zhang, Qian Zhang, and Shaofeng Platform," IJRECE, pp. 2166-2171, Vol. 6 Issue 3, July-September,
Lin. "Design of Security Gateway Based On Dual-Homed Architecture." 2018.
292 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Improvements in nanotechnology have introduced aware. Due to these variations, the solutions developed for
contemporary sensory devices that are capable of gathering WSNs cannot be directly applied to WMSNs. Therefore there
multimedia data in form of images, audio and video. Wireless is need to modify these techniques before they can be applied
multimedia sensor networks are designed to handle such type of to WMSNs. Furthermore, new techniques at all layers from
heterogeneous traffic. The ability to handle scalar and non-scalar physical layer to application layer suitable for these networks
data has led to the development of various real-time applications are required. Surveys on such research ranging from hardware
such as security surveillance, traffic monitoring and health to the network model layers and other cross-layer designs
systems. Since, these networks are an emergent of wireless sensor are[1], [4]–[6]. Some extensive studies on various hardware
networks; they inherit constraints that exist in these traditional
and software architecture test beds are in [7]. Transport
networks. Particularly, these networks suffer from quality of
service and energy efficiency due to the nature of traffic. This
protocols designed to be reliable are in [8]. A comparison of
paper presents the characteristics and requirements of wireless energy efficient and QoS aware routing protocols is done in
multimedia sensor networks and approaches to mitigate existing [9]–[11]. Accordingly, a review of QoS cognizant and multi-
challenges. Furthermore, a review of recent research on channel Media Access Control (MAC) protocols are in [12],
multipath routing protocols and multi-channel media access [13]. AlSkaif et al. present a comparative study on WSNs
protocols that have quality of service assurances and energy MAC protocols investigating their suitability on WMSNs
efficiency in handling multimedia data have included. through the analysis of some network parameters on node
energy drain [14]. Authors in [15], [16] identifies cross-layer
Keywords—Quality of service; multipath routing; multi- optimization solutions to problems inherent in WMSNs packet
channel media access control; energy efficiency delivery, energy preservation and error recovery. Discussions
of security requirements in WMSNs and classification of the
I. INTRODUCTION security threats as well as some protection mechanisms are in
Wireless Multimedia Sensor Networks (WMSNs) have [22], [23]. Finally,[19] discusses energy-efficiency issues with
enhanced the data gathering capability of the traditional regard to all sensor application designs as well as extension of
Wireless Sensor Networks (WSNs) which were restricted only network lifetime while [20] [21] proposes a classification of
to gathering scalar data. WMSNs have sensor nodes equipped energy-efficient target tracking schemes according to sensing
with cameras and microphones that enable these networks to and communication subsystems on a particular node.
gather multimedia data in various forms like live data streams,
This survey will thus concentrate on the important aspects
videos, audio, images and so on[1]. Recent advances in feature
required to deliver QoS-aware routing protocols in WMSNs,
engineering, image-processing techniques, machine learning
thus energy-efficiency, real-time multimedia streaming and
and communication technologies have given birth to various
data volumes. The paper will also highlight challenges and
research to applications of WMSNs. Applications include
proffered solutions to guide related research. Network
health care industry, military and general surveillance
designers and architects will also immensely benefit from the
systems, real time intelligent transportation systems and
clarity on characteristics and requirements of WMSNs as well
environmental monitoring [2], [3].
as existing solutions. Furthermore, presented is a survey of
The WMSNs are, descendent of WSNs hence the same communication MAC and routing protocols with emphasis on
benefits such as self-organization, flexibility, disposition energy-efficiency, scalability, QoS guarantee, prioritization
simplicity and scalability are also characteristic. However, the schemes, multipath routing and service differentiation
added features and capabilities in WMSNs present a number [17][18]. The conclusion will also give future directions on
of challenges that are inherent with these constrained networks discussed issues.
such as limited energy, storage, communication bandwidth as
The remaining paper is as follows: Section 2 highlights the
well as processing capacity. The large volumes of data
characteristics and design requirements of WMSNs with
generated by these multimedia networks to be reliably
design challenges and existing remedies. A study of multipath
transmitted over the wireless medium in real-time further
routing protocols is done in Section 3 followed by WMSNs
exacerbate these challenges. Research, on this domain, aims at
proposed MAC protocols. Lastly, Section 5 presents
the development of computation algorithms and protocols that
conclusions to the survey.
are highly energy-efficient and Quality of Service (QoS)
293 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
II. WIRELESS MULTIMEDIA SENSOR NETWORKS prudent that the communication protocols at every layer be
WMSNs are an emergent technology out of the traditional energy-efficient. For example, the transport layer protocols
WSNs. As such, they inherit many constrains that exist in reduce the number of control messages according to desired
these networks as well as new challenges and requirements levels of reliability, with routing protocols employing load
that come because of the requirement for real-time multimedia balancing and energy estimation techniques across the
services and handling of increased volumes of data. The network and at the MAC layer protocols can avoid idle
gathered data traffic handled by these networks requires listening by inactive nodes. Dynamic power management is
delivery in real-time due to the nature of applications that another important technique to be used as it ensures that idle
require the data. Examples of such applications include components of a sensor node are selectively shutdown or
security surveillance, health systems and traffic management hibernated to prevent unnecessary power consumption.
systems. The multimedia data collected by the camera sensors B. Real-Time Multimedia Data
is voluminous for a particular event; hence, bandwidth
In most applications involving multimedia data, QoS is
requirements for the transmissions are increased. As
difficult to achieve. Transmission of data to the sink without
summarized in Table I, WMSNs have opened many doors to
any packet loss or delays above a threshold is very crucial in
research due to their characteristics and capabilities. This
WMSNs. Therefore there is a need to impose severe QoS
section discusses the characteristics, design requirements of
demands on the networks. Delays cannot be tolerated in
WMSNs as well as proposed approaches [24].
applications that involve multimedia data for example in
A. Power Constraints security surveillance or traffic management systems. This
The camera sensor nodes in WMSNs are generally battery- implies that prioritization and service differentiation will play
powered. The batteries are expected to power the sensor nodes a pivotal role in these real-time systems. MAC protocols
for protracted periods without replacement. Therefore, the should give access or assign greater quality channels to higher
functionality of such nodes should take into cognizance these priority data. Routing protocols need to select paths that will
power constraints and limit energy consumption in its have the least delay to meet the required QoS as illustrated in.
computations and communication. In traditional WSNs, Reliability is also crucial in ensuring QoS to WMSNs.
energy drain due to computations can be insignificant Retransmissions are done at transport layer for example in
compared to WMSNs where computations tend to consume TCP while redundancy is at bit-level or at packet-level as
extremely high energy. To, capture and processing of a simple presented in [8]. However, these methods must be used with
frame in a vehicle tracking system can constitute up to 12% of consideration that they increase traffic hence consume more
total energy consumption of the overall event. It is therefore networks resources. The heterogeneous traffic in WMSNs that
recommended to adopt energy-efficient algorithms in image include multimedia and scalar data intended for different
processing [21] and likewise in video compression. Due to the applications with varying QoS demands will require variable
large volumes of multimedia data to be transmitted, it is levels of priority even within the same traffic type.
294 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
C. Volumes of Multimedia Data data efficiently. Mitigating the bandwidth constraint that is
extreme in WMSNs due to the large volumes and nature of
Typically, WMSNs have limited bandwidth hence traffic is also an important factor in achieving QoS
transmission of large volumes of sensory data presents a major communications. At the MAC layer, sensor nodes can
challenge to QoS guarantee. Techniques for data compression communicate simultaneously using different channels. Data
and redundancy reduction are vital to decrease data volumes traffic can be routed through multiple paths. However, radio
prior to transmission. One such technique is local processing equipment that has considerable bandwidth such as ultra-
where on-board analysis of the captured images is used to wideband (UWB) can be utilized in WMSNs.
extract only important events. The downside of local
processing is the requirement for added hardware resources. III. QUALITY OF SERVICE AWARE MULTIPATH ROUTING
Another technique is In-network processing of multimedia PROTOCOLS FOR WMSNS
data that encompass data fusion where the sink node collects
heterogeneous data from various nodes and create a Routing techniques for WSNs have been extensively
summarized version of events to reduce data redundancy and studied over the years to improve communications. However,
enhance inferences. To deal with the resource limitation these techniques cannot be directly implemented in WMSNs
problems associated with centrally coding data from multiple due to variations with traditional WSNs. Routing in WSNs
sensor cameras, WMSNs use distributed source coding (DSC) aims at finding the shortest path for transmission scalar data.
where encoding of data is done independently at each sensor Applying the same routing concepts to large volumes of
before transmission to the sink for decoding. This reduces the multimedia data will result in network congestions and
power consumption as well as required hardware resources. increased power drain on nodes. Therefore, the robust
Typically, WSNs transmit all collected data to the sink for approach will be to send data in parallel through multiple
subsequent processing and querying. Due to technological paths. Routing in WSNs is particularly concerned with
advancements, it is now possible to equip sensors with energy-efficiency whilst WMSNs also consider the QoS due
processors and flash memory that enable them to process and to real-time traffic and reliability concerns.
store data. After processing, only analyzed data is transmitted This section presents some multipath routing protocols in
to the sink. In terms of queries, only the result is sent to the WMSNs with QoS assurances. This survey looks at different
network after querying historical data. However, proper data protocols than those recently surveyed in. Furthermore, the
ageing schemes needs to be incorporated into the local chosen multipath routing protocols have single path routing
databases as they fill up in order to maintain data integrity. It support. For further comparison of the surveyed multipath
is also important to note that the sensors will form distributed routing protocols with QoS assurances, particularly to
databases which require efficient query engines to retrieve the WMSNs refer to Table II.
TABLE. II. COMPARISON OF MULTIPATH ROUTING PROTOCOLS UNDER REVIEW
295 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A multipath routing protocol based on ant colony coupling. However, the restriction is only to ZigBee tree
optimization called AntSensNet with QoS assurances is topology hence the paths are proportionate to available branches.
presented in. It has three phases of operation: Formation of the
M. Chen et al. recommended the directional geographical
cluster, route discovery phase, data transmission and route
routing (DGR) protocol for real-time video
maintenance. The cluster formation is initiated by the sink that
communications.The nodes in this protocol implement the
releases some cluster ants (CANTs). Those within close
global coordinate system to create virtual coordinates upon
proximity to the sink are selected as cluster heads (CH) and
receipt of a broadcast probe. The virtual coordinates are
will receive the CAs first. Upon receiving the CANTs, they
obtained by mapping the source and sink position along the x-
will be responsible for the reduction of the time-to-live (TTL).
axis to the destination or intermediate node. A node is selected
The cluster head will then advertise the CANTs to non-cluster
to be a forwarding candidate only if it falls within the
heads within its communication radius so that those who are
transmission range, the optimal mapping location and the
willing to join the cluster can join. Once clusters are formed,
threshold of the source. Next hop will be a candidate that has
the CH begins route discovery. Each CH manages a
the smallest distance to the optimal mapping hence; it will
pheromone table and shares with its neighbors according to
have a smaller timer than other competing nodes. If a timer
traffic classes following four parameters i.e. Energy, packet
expires, the node sends a reply message REP to the source. On
drop, memory and delay. Traffic specific paths to the sink is
receipt of an REP, the source confirms with SEL message.
created by broadcasting a forward ant (FANT) which will
Nodes that hear the REP or SEL cancels their timers. The
collect traversed node identities and the four parameters
winner node will not establish any other path to the same
(queue delay, ratio of packet, residual energy and available
source in order to guarantee path disjointedness. In turn, the
memory) as it propagates. When a node receives a FANT, an
connected node will send its own probing messages following
update is done to its information before sending it to the next
the same procedure with an adjusted deviation angle to create
hope that satisfies the QoS requirements and a corresponding
a path towards the sink. For establishment of multiple paths,
backward ant (BANT) is transmitted in the reverse path for
the source will send a number of probe messages with
path reservation. On receipt of the BANT, nodes update their
variations in the initial deviation angle. For video routing, the
pheromone tables. For establishment of multiple paths for
source broadcasts the complete frame initially to all single hop
video transmission, a video forward ant (VFANT) is
neighbors. Those neighbors within the chosen paths will
broadcasted in the same manner as the FANT and the sink
retransmit the video using respective paths only those packets
responds by sending multiple VBANTs. The VBANTs will be
specified by the source. The packet delivery in this protocol is
used to choose paths for sending video data. Once routes are
fast and reliable through multipath and the forwarding
ready data, delivery starts. A maintenance ant (MANT) is used
equivalence class. It also scales well due to the stateless
for route maintenance. This protocol gives differentiated
geographic based routing paradigm. However, if a node fails,
service to ensure QoS delivery by offering each traffic
the path recovery takes longer as well as the new route
separate routes. The use of cluster heads is a drawback on
discovery. In addition, it considers only a single active source
scalability. However, the multipath routing technique is viable
for video transmissions that might not be practical in some
for video data only.
scenarios.
Z. Bidai et al. proposed the ZigBee Multipath Hierarchical
A. Bhattacharya and K. Sinha following the principles of
Tree Routing (Z-MHTR) protocol. It allows source to use non-
ad-hoc on-demand distance vector routing (AODV) developed
parent neighbors to search for other paths. The source node
the least common multiple routing (LCMR) protocol. As
maintains a record of all branches used for tree routing (TR).
opposed to calculating the shortest path by number of hops, it
The source node will construct disjoint paths using three basic
uses the routing time taken or end-to-end delay to choose
principles. If a selected next hop node branch has not been
multiple paths. During route discovery, the route reply
utilized for TR path by the source node then a node disjoint
message RREP has to arrive before the deadline otherwise it
can be established from that node to the sink using TR. If the
will not be accepted. The source node uses the RREP message
branch has already been utilized for TR path by the source
to check the routing time taken by the corresponding route
then the next hop will depend upon the depth of a node
request message RREQ before reaching the destination. From
common to the TR path used by the source and the node that
the accepted x paths that have routing time {T1, T2, …,Tx}, it
has used the node branch for TR. If all neighbor’s branches
calculates the least common multiple L of {T1 ,T2, … ,Tx}. The
have been utilized in TR then it selects the neighbor node that
packets sent over path i are decided such that = ∑
is not in any TR path. The rules are applied to any subsequent
packets, L/Ti packets will be routed along that path i. The total
nodes until the sink. The number of disjoint paths corresponds
time it takes to deliver k packets gives the maximum routing
to the number of branches forming the topology. Furthermore,
time Tmaxof {T1 ,T2, … ,Tx}. This protocol ensures avoidance of
the author proposed for reduction of interference in which
congested routes through the end-to-end calculation of routing
nodes lists interfering neighbors except the ones on the same
time during its route discovery process. In order to reduce the
paths. This is done by checking whether they can hear data
transmission time, the number of packets allotted to a
packets that are not destined to them. The disjoint paths that
particular route is reduced according to time L and the routing
reduce inter-path interloping are preferred. Based on the
time Ti of the path. However, this may lead to early node death
ZigBee tree topology and address assignment, multipath
if most traffic is continuously routed through a node with least
routing is achieved through neighbor table and a record of
end-to-end routing time. Adaption to congestion and route
routing tree usage on a particular branch. The further work
breakage needs improvement.
mitigates multiple paths interferences caused by route
296 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Unlike DGR, that uses the deviation angle for controlling IV. QUALITY OF SERVICE AWARE MEDIA ACCESS CONTROL
the directions of multiple paths, Li et al proposed the division PROTOCOLS FOR WMSNS
of the topology into different districts for specific paths using
MAC protocols present a challenge during their design and
the geographic energy-aware non-interfering multipath routing
implementation when aiming for energy efficiency and
(GEAM) protocol. After division into virtual coordinates just
coordinating transmission of large volumes of multimedia
like in DGR, the source and sink areas are restricted within the
sensory data and meeting QoS in MWSNs. The dynamic and
transmission radius. Each packet is piggybacked with
burst traffic predominant in WMSNs it requires application of
boundary information of the selected district by the source
duty cycling techniques in saving energy deeper analysis.
before transmission. The subsequent nodes will then use
Reduction of collisions is also an important factor in MAC
greedy perimeter stateless routing (GPSR) to forward the
protocol design especially when it involves real-time
packet to the respective district. For load balancing and even
multimedia data. Controlling media access through
distribution of energy, GEAM the data transmissions are
prioritization and differentiation of services is also an
organized in runs of same lengths. To further avoid
important factor when handling heterogeneous traffic. This
interference within multiple routing paths, it applies division
section will elaborate some of the energy-efficient MAC
of runs into three rounds, where a district D x belongs to round
protocols that have QoS assurances. A summary of the same is
k if Dx%3 = k. During the first run, loads are distributed
in Table III.
evenly to all districts. After each run, the sink collects residual
energy from all nodes within a district and sends back to the M. Arifuzzaman et al. proposed the intelligent hybrid
source. Based on these statistics the source adjusts the rate of MAC (IH-MAC) protocol. This protocol combines CSMA/CA
utilization for every district and those with higher energy and TDMA techniques as a single mechanism that implements
levels get more loads in the next run. GEAM achieves local synchronization. The protocol prioritizes the node
balanced traffic loads and energy consumption as well as holding data with high QoS such as real-time data. If nodes
avoids interference by the division to the topology into various have same priority and mapped to same slot, then they
districts. Scalability is also guaranteed using GPSR. However, contend for that slot. For energy preservation, it adjusts its
piggybacking every packet with border information and transmission output during the contentions. The protocol
making it collect network statistics increases the overhead. It scales well and reduces collisions as well as improves on
also does not consider some QoS metrics such as delay and channel utilization and access delays that are challenges in
reliability that are of paramount importance to delivery of CSMA/CA by fusion of CSMA/CA and TDMA.
multimedia data.
TABLE. III. COMPARISON OF MEDIA ACCESS CONTROL PROTOCOLS
traffic types,
Prioritization traffic types, traffic types,
traffic types traffic types traversed hop traffic types traffic types
Scheme packet lifetime dynamic
count of packets
dynamic channel
adaptive
adaptive allocation, adaptive adaptive
contention
Service contention dynamic slot contention contention
dynamic slot window, dynamic channel
Differentiation window, allocation, window, window,
allocation dynamic duty access control
Scheme dynamic duty adaptive dynamic slot dynamic slot
cycle, weighted
cycle contention allocation allocation
fair queueing
window
Scalability poor good good medium medium poor poor
Adaptation to
good medium medium poor good good poor
Dynamic Traffic
Collision Rate low medium medium low low low high
Energy Efficiency good medium medium good medium poor good
Message Passing no no yes no no yes no
Clustered yes no no yes yes no no
Year 2008 2008 2011 2015 2013 2014 2016
297 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
An energy-efficient hybrid MAC scheme (EQ-MAC), was weighted fair queuing (WFQ) technique to allow channel
proposed by Yahya and Ben-Othman. It uses the cluster access to real-time traffic while Saxena et al. uses a FIFO
mechanism in which the cluster head schedules slots using mechanism. Diff-MAC avoids starvation to same traffic type
TDMA. It uses frames for communication. The cluster head by prioritization of packets belonging to the same queue
sends the initial broadcast frame for synchronization. Once prioritizing them based on traversed hops. It further segments
synchronization is completed, the cluster members start video frames and transmit the in bursts to lower
transmission of data through the cluster head. The cluster head retransmission cost. Both protocols use the dynamic duty
issues TDMA slots upon request from the cluster members cycle technique. The protocols offer good QoS, fairness and
with consideration of traffic priorities. The cluster head then energy-efficiency in WMSNs. However, constantly
broadcasts allocated TDMA slots to cluster members for monitoring of various states in a network leads to idle
transmissions to begin. Sleep mechanism will also apply to listening and as for Diff-MAC, the constant intra-queue
those cluster members without data to transmit. Real-time data prioritizations may not scale well with high traffic.
is placed in a queue that is served instantaneously. The sleep
mechanism saves energy and channel utilization. The protocol MQ-MAC is a cluster based slotted CSMA/ CA MAC
assures delivery of real-time data especially multimedia due to protocol. The cluster head is responsible for key
prioritization of traffic. However, this may starve low priority responsibilities that include channel sensing, time slot
traffic. allotments and channel allocation. It divides its super frame
into active and sleep periods, with the active being sub-
An efficient QoS provisioning protocol by M. Souil divided into three phases namely; sensing, channel selection
(AMPH), is a hybrid channel access method. The notable and data transmission requests. Once the cluster head receives
difference between AMPH and IH-MAC is that the latter is results of channel sensing and transmission requests from the
CSMA/CA centered and AMPH is TDMA centered. AMPH cluster members, it will allocate slots and transmission
divides transmissions into slots and two-hop radius for each channels. QoS is guaranteed through slot allocation. The
node. Prioritization for medium access is done by separation requests once received from cluster members are classified
of real-time and best effort traffic and based on slot according to arrival time and traffic type as well as
ownership. Contending nodes are separated into four groups consideration of the packet lifetime. Early slots are allocated
according to traffic priority: real-time by owner, real-time by to requests with higher priority. The slot allocations are allows
non-owner, best effort by owner and best effort by non-owner. data traffic from cluster members to the cluster head to be
To avoid starvation, the protocol allows best effort traffic collision free. After the transmission phase, the sensor nodes
ahead of real-time traffic in limited slots per cycle. To will sleep and wake up when another super frame starts. QoS
conserve energy, it allows nodes to switch of their radios in is guaranteed through allocation of slots and channels for
the waiting state. The use of any slot coupled with traffic different traffic types according to priority. However, the
prioritization achieves optimum channel utilization and QoS presence of many control messages during sensing and
guarantees to heterogeneous traffic. However, there is need for switching are not desirable due to overheads.
a robust differentiation of traffic that caters for more traffic
types that exist in WMSNs. V. CONCLUSION
Bhandari et al. proposed a multi-channel priority based WMSNs are becoming more popular in various IoT
adaptive MAC protocol (PA-MAC) that is based on the IEEE applications due to their ability to handle heterogeneous data
802.15.4 standard. The protocol traffic classification is from various sensory devices. Considerable research has been
grouped into four categories according to priority: emergency done to enhance these networks. However, some challenges
(medical), on-demand, normal, non-medical. It uses the are still prevalent due to the distinctive characteristics of the
contention access periods (CAP) for the four classifications of WMSN and resource constraints. This paper covered the
traffic. Traffic with higher priority is allowed access to slots unique characteristics and requirements for WMSNs as well as
for lower priority traffic and the lower priority traffic some design approaches to the constraints highlighted.
transmits during the contention free period (CFP). The nodes Furthermore, the survey includes multipath routing protocols
enter into sleep until next transmission. Collisions are and MAC protocols, which are two important communication
mitigated by traffic differentiation and transmission of lower parameters to improve QoS provision in any network.
priority data (e.g. multimedia data in medical scenario) at Multipath routing is significant to the provision of QoS and
CFP. However, the protocol gives less priority to multimedia delivery of multimedia data in WMSNs. These protocols are
data hence it cannot be applied directly to WMSNs. able to distribute the voluminous multimedia data across the
network, thus balancing the load as well as energy
Related CSMA/ CA based protocols with QoS assurances consumption. It is important for the protocols to counter
were proposed by Saxena et al. and Diff-MAC. The protocols interference in multiple parallel paths to avoid route coupling
use adaptive contention window (CW) and dynamic duty issues. However, most multipath routing protocols consider
cycling mechanisms. The CW sizes for real-time traffic are set load balancing and energy management without due diligence
to be less than low priority traffic. The protocols differ in that, for other QoS metrics such as prioritization, and
Saxena et al. aims for fairness by making sensors adjust their differentiation of service. Traffic in these networks is
CW size after checking with neighboring sensors if chances of heterogeneous in nature therefore prioritization and service
a collision remain after last CW size changes whereas sensors differentiation should not only be fixed to a particular type of
in Diff-MAC continue to change their CW sizes towards the traffic as in the case with most protocols that dedicate only to
threshold CW size. Diff-MAC also employs the hybrid video traffic. Route recovery and congestion control must also
298 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
be given great significance to improve QoS in WMSNs. [11] S. Ehsan and B. Hamdaoui, “A Survey on Energy-Efficient Routing
Finally, Efficient MAC protocols intended for WMSNs must Techniques with QoS Assurances for Wireless Multimedia Sensor
Networks,” IEEE Commun. Surv. Tutorials, vol. 14, no. 2, pp. 265–278,
be able to handle heterogeneous traffic and vast volumes of 2012.
multimedia data. In literature, there exist CSMA/ CA based [12] O. D. Incel, “A survey on multi-channel communication in wireless
MAC protocols that are scalable and adapt to different sensor networks,” Comput. Networks, vol. 55, no. 13, pp. 3081–3099,
variable traffic situations although suffer bottlenecks in QoS Sep. 2011.
provision and energy efficiency. Hybrid protocols combining [13] M. A. Yigitel, O. D. Incel, and C. Ersoy, “QoS-aware MAC protocols
CSMA/ CA and TDMA are promising to be an important part for wireless sensor networks: A survey,” Comput. Networks, vol. 55, no.
of WMSNs since CSMA/ CA and TDMA are used to handle 8, pp. 1982–2004, 2011.
low data rates and high data rates respectively thereby [14] T. AlSkaif, B. Bellalta, M. G. Zapata, and J. M. BarceloOrdinas,
“Energy efficiency of MAC protocols in low data rate wireless
improving throughput and the reduction of collisions. It is multimedia sensor networks: A comparative study,” Ad Hoc Networks,
important to note that QoS is of great importance to WMSNs vol. 56, pp. 141–157, Mar. 2017.
hence the future research focus should be on handling [15] D. G. Costa and L. A. Guedes, “A survey on multimedia-based cross-
multimedia data collected by camera sensors. layer optimization in visual sensor networks,” Sensors, vol. 11, no. 5,
pp. 5439–5468, 2011.
REFERENCES
[16] N.Thangadurai, Dr.R.Dhanasekaran and R.D.Karthika, "Dynamic
[1] T. Almalkawi, M. G. Zapata, J. N. al-Karaki, and J. Morillo-Pozo,
Energy Efficient Topology for Wireless Ad hoc Sensor Networks",
“Wireless multimedia sensor networks: Current trends and future
WSEAS Transactions on Communications, Vol. 12, Iss. 12, pp. 651-
directions,” Sensors, vol. 10, no. 7, pp. 6662–6717, 2010.
660, 2013.
[2] T. Semertzidis, K. Dimitropoulos, A. Koutsia, and N. Grammalidis,
[17] L. D. P. Mendes and J. J. P. C. Rodrigues, “A survey on cross-layer
“Video sensor network for real-time traffic monitoring and
solutions for wireless sensor networks,” J. Netw. Comput. Appl., vol.
surveillance,” IET Intell. Transp. Syst., vol. 4, no. 2, p. 103, 2010.
34, no. 2, pp. 523–534, 2011.
[3] N. B. Bo et al., “Human mobility monitoring in very low resolution
[18] N. Thangadurai and Dr.R.Dhanasekaran, “Energy Efficient Cluster
visual sensor network,” Sensors (Switzerland), vol. 14, no. 11, pp.
based Routing Protocol for Wireless Sensor Networks”, International
20800–20824, 2014.
Journal of Computer Applications, Vol. 71, No. 7, pp. 43-48, 2013.
[4] S. Soro and W. Heinzelman, “A Survey of Visual Sensor Networks,” [19] N.Thangadurai, Dr.R.Dhanasekaran and R.D.Karthika, "Dynamic
Adv. Multimed., vol. 2009, pp. 1–21, 2009.
Traffic Energy Efficient Topology based Routing Protocol for Wireless
[5] A. Sharif, V. Potdar, and E. Chang, “Wireless multimedia sensor Ad hoc Sensor Networks", International Review on Computers and
network technology: A survey,” IEEE Int. Conf. Ind. Informatics, no. Software, Vol. 8, No. 5, pp. 1141-1148, 2013.
May 2014, pp. 606–613, 2009.
[20] T. Winkler and B. Rinner, “Security and Privacy Protection in Visual
[6] I. F. Akyildiz, T. Melodia, and K. R. Chowdury, “Wireless multimedia Sensor Networks,” ACM Comput. Surv., vol. 47, no. 1, pp. 1–42, 2014.
sensor networks: A survey,” IEEE Wirel. Commun., vol. 14, no. 6, pp.
[21] M. Guerrero-Zapata, R. Zilan, J. M. Barceló-Ordinas, K. Bicakci, and B.
32–39, 2007.
Tavli, “The future of security in Wireless Multimedia Sensor Networks :
[7] I. F. Akyildiz, T. Melodia, and K. R. Chowdhury, “Wireless multimedia A position paper,” Telecommun. Syst., vol. 45, no. 1, pp. 77–91, 2010.
sensor networks: applications and testbeds,” Proc. IEEE, vol. 96, no. 10,
[22] T. Rault, A. Bouabdallah, and Y. Challal, “Energy efficiency in wireless
pp. 1588–1605, 2008.
sensor networks: A top-down survey,” Comput. Networks, vol. 67, no.
[8] M. A. Mahmood, W. K. G. Seah, and I. Welch, “Reliability in wireless March, pp. 104–122, 2014.
sensor networks: A survey and challenges ahead,” Comput. Networks,
[23] O. Demigha, W. K. Hidouci, and T. Ahmed, “On Energy efficiency in
vol. 79, pp. 166–187, 2015.
collaborative target tracking in wireless sensor network: A review,”
[9] M. Radi, B. Dezfouli, K. A. Bakar, and M. Lee, “Multipath routing in IEEE Commun. Surv. Tutorials, vol. 15, no. 3, pp. 1210–1222, 2013.
wireless sensor networks: Survey and research challenges,” Sensors, vol.
[24] S. Bhandari and S. Moh, “A priority-based adaptive MAC protocol for
12, no. 1, pp. 650–685, 2012.
wireless body area networks,” Sensors (Switzerland), vol. 16, no. 3,
[10] A. M. Zungeru, L.-M. Ang, and K. P. Seng, “Classical and swarm 2016.
intelligence based routing protocols for wireless sensor networks: A
survey and comparison,” J. Netw. Comput. Appl., vol. 35, no. 5, pp.
1508–1536, Sep. 2012.
299 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The real time vehicle movement traces using surface in Geostationary Orbit (GEO) at 83° East, 32.5° East,
waypoint display on the base-map with IRNSS/NavIC and GPS and 131.5° East longitude. The movement of four GSO
dataset in the GUI simultaneously. In this paper, a portable satellites will be in the form of figure of “8” as shown in
electronic device with application software has been designed and Fig. 1.
developed, which would be used to capture the real time
positional information of a rover using IRNSS-UR. It stores the Ground segment: The maintenance and operation of the
positional information into database and displays the real time IRNSS constellation is held in ground segment (Fig. 2) are:
vehicle positional information like date, time, latitude, longitude
and altitude using both GPS and IRNSS/NavIC receiver 1) IRNSS Spacecraft Control Facility (IRSCF)
simultaneously. The designed hardware device with an 2) ISRO Navigation Centre (INC)
application software developed helps in mapping the real time 3) IRNSS Range and Integrity Monitoring Stations
vehicle / rover movement at the same time which also helps in (IRIMS)
identifying the region with data loss, varying positional 4) IRNSS Network Timing Centre (IRNWT)
information, comparing the distance travelled by rover and also 5) IRNSS CDMA Ranging Stations (IRCDR)
aid in retrieving the past surveys and mapping the traces of both
IRNSS and GPS simultaneously. The vehicle movement using
6) Laser Ranging Stations
both IRNSS/NavIC and GPS are tracked on the base map to find 7) IRNSS Data Communication Network (IRDCN)
the similarity and differences between two. During this research CDMA extending is being completed by the four IRCDR
work it can be conclude that that the rover position using GPS stations all the time for all the NavIC satellites. The INC built
and IRNSS were accurate and continuous in our survey duration
up at Byalalu performs remote tasks and information
except in few places. In that few places the data loss is observed
because of the satellite visibility variations. For Indian region the
accumulation with all the ground stations. The IRNWT has
IRNSS/NavIC can be a better replacement for GPS. been set up and is furnishing IRNSS framework time with an
exactness of 2 ns (2.0×10−9 s) (2 sigma) with respect to UTC.
Keywords—GNSS; GPS; IRNSS; embedded systems 14 IRIMS are right now operational and are supporting NavIC
operations. Laser extending is being completed with the help
I. INTRODUCTION of ILRS stations the world over. Route Software is operational
NavIC/IRNSS is a free provincial regional satellite at INC since 1 August 2013. The IRDCN has built up earthly
constellation being developed by India. IRNSS will give two and VSAT connects between the ground stations. Seven 7.2 m
kinds of administration services, to be specific, Standard FCA and two 11 m FMA of IRSCF is right now operational
Positioning Service (SPS) which is given to every one of the for LEOP and on-circle periods of IRNSS satellites [3]. All
customers and Restricted Service (RS), which is an encoded the route parameters viz. satellite ephemeris, clock revisions,
administration given distinctly to the approved clients only. It trustworthiness parameters and auxiliary parameters viz. iono-
is intended to give precise position data administration to postpone remedies, time counterbalances with respect to UTC
customers in India just as the locale extending out up to 1500 and different GNSS, chronological registry, instant message
km from its limit, which is its essential administration region. and earth direction parameters are created and uplinked to the
An all-inclusive administration territory lies between essential shuttle naturally [1].
administration region from Latitude 30 0 S to 500 N and Signal: Standard Positioning Service are modulated by a 1
Longitude 30 degree East to 130 degree East [2]. The IRNSS MHz BPSK signal and a Precision Service will use BOC (5,2)
system is relied upon to give a position precision of superior are the two NavIC signals consists of S1 band (2492.028
to 10 m in the essential administration region [5]. MHz) and L5 (1176.45 MHz). An informing or data interface
The NavIC system comprises a ground segment which is is implanted in the NavIC framework. The navigation signals
supported by a space segment. themselves would be transmit in the S band recurrence (2–4
GHz) and communicate through a staged exhibit reception
Space segment: 8 satellites constellation, where 4 satellites apparatus i.e., antenna to keep up required inclusion and signal
are in Geosynchronous Orbit (GSO) crossing the equator at quality. This element permits to send admonitions to a
55° East and two at 111.75° East and three satellites are particular geographic region. For instance, fishermen utilizing
located approximately 36,000 km (22,000 mi) above earth the framework can be cautioned about a cyclone [3].
300 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 2. IRNSS Operation Model [ISRO]. Fig. 3. Antenna Setup in Mobile Vehicle.
Accuracy: The framework is expected to give a complete The hardware device which basically consists of
position exactness of superior to anything 10 meters all Raspberry pi board and other peripherals have been developed
through Indian landmass and superior to anything 20 meters in an application in order to read NMEA (National Marine
the Indian Ocean just as an area broadening roughly 1,500 km Engineering Association) data from the IRNSS-UR at 1 Hz.
(930 mi) around India. NavIC has dual frequencies i.e., S1 and
L5 bands whereas GPS which is dependent on L band only. At The application is programmed to read the NMEA from
the point when low-recurrence sign goes through the climate, the IRNSS-UR receiver, parsed the NMEA data, storage and
its speed changes because of air unsettling influences. For display colored marker on GUI to indicate the current vehicle
India's situation, the real delay is evaluated by estimating the position. The hardware and application has been tested during
distinct differences in two frequency i.e., L and S band. this survey and noted the bugs / issue / short comings of both
Thusly, NavIC isn't subject to any model to discover the hardware and software.
recurrence mistake and is more precise or near to the GPS
Few of the bugs encountered during this survey are:
[6][8][10].
1) The device was able to read data but the delay for data
Major applications of NavIC are: Disaster Management,
Terrestrial, Precise Timing, Aerial and Marine Navigation, reading was incremental and that leads to incremental time
Mapping, Vehicle tracking, Integration with mobile phones, differences between current time and time read by the device.
Voice and visual navigation for drivers and fleet management, 2) When there is switch between poor signals strength to
Terrestrial navigation aid for hikers and travelers, and good signal strength (vice-versa) the parsing of NMEA data
Geodetic data capture. was incorrect.
There are various challenges involved in replacing the Two separate applications developed to fix the problem of
existing GPS navigation system with autonomous regional -“NMEA data reading at incremental delay which causing loss
navigation system. In this paper aims at plotting or mapping of data”.
the rover position in real time using both GPS and NavIC
receivers. The purpose of real time mapping of rover position 1) One of the applications is meant for reading NMEA
is to identify the geographical region with similarity and data, saving data in separate text file with auto count. The rate
differences between GPS and NavIC navigational systems [7] of creating the text file is fixed at the rate of every 30 seconds
[9]. (new text file / 30 sec).
301 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
302 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
303 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The Ethernet adapter is internally connected to an a) To view the NMEA data in the GUI: Connect RS 232
additional USB port. In Model A, A+, and the Pi Zero, the cable from Laptop to the NMEA/LCD port of receiver through
USB port is connected directly to the System on Chip (SoC). the USB-to RS-232 converter (Table III).
On the Pi 1 Model B+ and later models the USB/Ethernet chip b) To collect IF samples into PC: Connect USB cables
contains a five-point USB hub, of which four ports are from Laptop to the IF SAMPLES TO PC port and SD Card
available, while the Pi 1 Model B only provides two. On the Pi
Data TO PC port of receiver.
Zero, the USB port is also connected directly to the SoC, but it
uses a micro USB (OTG) port. c) To collect IF samples into Hard-disk: Connect USB
cable from External Hard disk to the IF SAMPLES TO HARD
A typical block diagram of the IRNSS-UR is shown in Fig. DISK port of receiver.
8. IRNSS-UR is expected to receive, down convert and
d) To run the Receiver with External clock source: For
demodulate the transmitted satellite signals both at L5
(1176.45 MHz) and S1 (2492.028 MHz) band frequencies. External clock source, connect SMA Cable from external
Most importantly, IRNSS-UR generates measurements clock source to the 10 MHz IN port of receiver.
precisely with respect to the external/internal trigger such as 1 e) To Latch the Receiver measurements with external
Pulse per Second (PPS). In addition, IRNSS-UR shall also PPS: Connect SMA Cable from external PPS source to the
include capability to process Global Positioning System (GPS) EXT PPS IN port of receiver.
L1 Coarse/Acquisition (C/A) signals centered at 1575.42 MHz f) To log the Navigation data of IRNSS into SD card:
and generate measurements with respect to the Insert SD card into the SD CARD slot of receiver.
external/internal trigger. The receiver shall output the user
position computed using L5 only, S1 only; L1 only, combined
L5 and GPS and combined S1 and GPS and combined IRNSS
and GPS. Control inputs to the system include 10 MHz
external reference clock, 1-PPS signal and commands/data
through Ethernet and/or RS232 interface. The IRNSS-UR
shall also have an USB port for IF sampled data collection
into the PC and an external storage device.
B. Antenna Setup
Antenna module contains a mounting interface and an RF
connector on bottom plate. Mounting interface is a standard
M16 nut and can be mounted on standard M16 pipe/threaded
pole (Fig. 9) [4]. Ensure there are no metal parts nearby the
antenna for optimistic performance during installation. RF
connector is a TNC Female type, through which the RF signal
received by the antenna is fed to the receiver and the DC
supply from the receiver is fed to the antenna LNA.
1) The antenna mounting site should provide full 360-
degree visibility of the horizon. Any physical obstruction
Fig. 8. High Level Block Diagram of IRNSS-UR.
having an apex that makes an angle more than 5 degrees with
the antenna phase centre, degrades the unit performance by
blocking the satellite signals.
2) Ensure that there is no metal objects/plates touching or
very close to the antenna. This is because the metal object
alters the gain pattern of the antenna.
C. Receiver Setup
The back panel of the IRNSS-UR is as shown in the Fig.
10(a). The details and the functionalities of the ports in the
back panel of the IRNSS-UR are as mentioned in Table I and
the front panel of the IRNSS-UR is as shown in Fig. 10(b).
The details and the functionalities of the ports in the front
panel of the IRNSS-UR are as mentioned in Table II.
The steps to be followed depending on the Users’ data
requirement in a particular signal/data output are (Fig. 11): Fig. 9. IRNSS Antenna.
304 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
305 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
306 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
307 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
This device can be further developed to guide and monitor [3] Mohmad Umair Bagali, Dr. Thangadurai N, “Embedded Board
the all transportation vehicles including aircrafts and keep Development Interfaced with GPS/IRNSS/NavIC Receiver for Disaster
Applications”, Proc. Of SSRN, International Conference on Sustainable
them informed about the routes of the neighboring aircrafts in Computing in Science, Technology & Management, pp. 416 – 426,
the airspace. Jaipur, 2019.
[4] Mohmad Umair Bagali, Naveen Kumar Reddy, Ryan Dias, Dr.
ACKNOWLEDGMENT Thangadurai N. The Positioning and Navigation System on Latitude and
This research work was carried out by the grant sanctioned Longitude Map using IRNSS User Receiver. IEEE International
Conference on Advanced Communication Control and Computing
by ISRO File No. NGP22. The authors like to acknowledge Technologies, Ramanathapuram, pp.122-127, 2016.
the support of Space Applications Centre - Indian Space [5] Rao, V. G., Lachapelle, G. and Vijay Kumar, S. B., Analysis of IRNSS
Research Organization (SAC-ISRO) by providing IRNSS over Indian Subcontinent. J. Inst. Navigation, San Diego, 2011.
receiver and encouraging us to work on this navigation [6] Kaplan, “Understanding GPS: principles and applications”, 2nd Ed,
studies. Authors thank Dr. G. Raju & JAIN (Deemed-to-be Artech House, 2006.
University), Bangalore for the support to complete this work [7] Ganeshan, A.S., Rathnakara S.C., Gupta R., and Jain, A.K., Indian
successfully. Scholar also acknowledges the Directorate of Regional Navigation Satellite System (IRNSS) Concept, ISRO Satellite
Minorities, Government of Karnataka for their Ph.D. Center Journal of Spacecraft Technology, 2005, 15(2), pp. 19–23.
Fellowship. [8] Mohmad Umair Bagali, Thangadurai N, “Application Specific
Embedded Board Development Interfaced with GPS/IRNSS Receiver
REFERENCES for Environmental Monitoring”, International Journal of Innovative
[1] Chandrasekar, M. V. et al., Modernized IRNSS broadcast ephemeris Technology and Exploring Engineering, Vol.8, Iss.8, pp. 2628–
parameters. J. Control Theory Inf., vol. 5, Iss.2, 2015. 2637,2019.
[2] ISRO, Indian Regional Navigation Satellite System Signal in Space ICD [9] Grewal, M.S., Weill, L.R., and Andrews, A.P., Global Positioning
for Standard Positioning Service (Version 1.0, ISRO-IRNSSICD- SPS- Systems, Inertial Navigation and Integration, Wiley Publications, 2007,
1.0), Indian Space Research Organization, 2014. 2nd edition.
[10] Van Diggelen, F., GNSS Accuracy: Lies, Damn Lies, and
Statistics, GPS world, 2007, pp. 26–32.
308 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The classification and prediction of medical knowledge discovery is the name of the discovery of hidden
diseases is a cutting edge research problem in the medical field. knowledge from large databases. Knowledge discovery
The experts of machine learning are continuously proposing new contains the steps of data preparation, data preprocessing, and
classification methods for the prediction of diseases. The hypothesis generation, the formation of the pattern, evaluation,
discovery of classification rules from medical databases for knowledge representation, knowledge refinement, and
classification and prediction of diseases is a challenging and non- knowledge management [2]. It also includes many stages for
trivial task. It is very significant to investigate the more databases updating.
promising and efficient classification approaches for the
discovery of classification rules from the medical databases. This Machine Learning methods and biological databases play a
paper focuses on the problem of selection of more efficient, significant role in disease diagnosis. It helps in future for
promising and suitable classifier for the prediction of specific diagnosing of medicine. The biological database includes
diseases by performing empirical studies on bunch mark medical information about gene structure, function, and similarities of
databases. The research work under the focus concentrates on structure and sequences of biological data. Classification of the
the benchmark medical data sets i.e. arrhythmia, breast-cancer, biological database can be done in two forms as a specialized
diabetes, hepatitis, mammography, lymph, liver-disorders, sick, and comprehensive database. The comprehensive database
cardiotocography, heart-statlog, breast-w, and lung-cancer. The includes different species database, for example, GenBank [3]
medical data sets are obtained from the open-source UCI and specialized databases consist of a special organism or
machine learning repository. The research work will be
species databases, for example, WormBase [4].
investigating the performance of Decision Tree (i.e.
AdaBoost.NC, C45-C, CART, and ID3-C) and Support Vector Machine learning becomes a necessary part of solving the
Machines. For experimentation, Knowledge Extraction based on problem in every branch of science. In biomedicine to predict
Evolutionary Learning (KEEL), a data mining tool will be used. genetic sequence and protein structure machine learning has
This research work provides the empirical performance analysis been used [5]. Machine learning is used to extract hidden
of decision tree-based classifiers and SVM on a specific dataset. knowledge for the different data set. It includes neural network,
Moreover, this article provides a comparative performance boosting, support vector machine and decision trees [6]. In
analysis of classification approaches in terms of statistics.
machine learning, two ways are performed for data mining. It
Keywords—Classification; rules discovery; support vector
is supervised learning we make a dataset to extract new data
machine; decision tree from a large amount of data. New data and training data set
match for validation of result. But in unsupervised learning,
I. INTRODUCTION some pattern is used to classifying the data without explicit
instruction [7]. Reinforcement learning focus on the reward
The Knowledge Discovery is processing of finding the non-
and output achieve in the form of reward and punishment. An
trivial, useful and hidden patterns from a very large database.
agent is required to gain the maximum reward to gain the
Knowledge discovery and data mining are a new trend in
result. Agent focuses on the positive situation to gain
information technology. Traditionally a large part of the
maximum reward. Negative situation decreases the reward.
process was done by manually that is time-consuming task.
This type of learning is used in control theory, statistics,
With time new technologies invented and task shifted from
information theory, etc.
manually to computerized form. Business knowledge is
necessary in advance to compete in the world. Data storage is This research article investigates the performance of
now a day reached to amount of terabyte size [1]. But it is Decision Tree approach and Support Vector Machine
necessary to extract useful knowledge from it for use. So Algorithm for the discovery of classification rules. The
*Corresponding Authors.
309 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
interesting and useful discovered rules are used for the building classification algorithm and accuracies of classification are
of classifiers. The classifiers are applied for the diagnoses of performed by 10 cross validation techniques. By using
the various harmful diseases. In this paper, we use KEEL [8] different classifier they concluded that Random forest takes
data mining tool for the data processing and classification of less running time with the highest accuracy of 87.50%. This
the biological databases. accuracy gives help in ailment prediction and classification in
the field of medical science [12].
Section II provides the related work published in
contemporary literature. Section III gives information about the In this study extended deep learning method is used for
decision tree-based classification and provides the empirical classifying multimedia data set. Convolution Neural Network
performance analysis of selected classifiers on medical is a deep learning method is costly but this paper feed low level
databases. Section IV provides a basic understanding of SVMs features in this approach. To find the best result CNN is used
and comparative empirical study on medical data sets. with the bootstrapping method. TRECVID data set is used in
Sections V and VI provide the experimental setup and this approach which is high-level imbalanced data set. This
discussion on the results produced during the under focused approach works effectively on the use of low-level features that
research study and the last section concludes the findings of the reduced the training time in deep learning [13].
research work.
Anuj et al. describe Parkinson's disease. It is the connection
II. RELATED WORK between speech impairment and Parkinson's disease. In this
paper classification based deep learning (Deep Neural
This section provides the literature review of the various Network, Dimensionality reduction techniques) and machine
research carried by the different researchers in this field. The learning algorithms (Logistic regression, Naïve Bayes, K-
following section gives information about the use of different Nearest Neighbor, Decision tree, Random forest) are employed
classification for the discovery of rules and the classification of with the use of Dimensionality reduction. The data set
different biological diseases. Parkinson's Speech is used in this approach that is obtained
There are many techniques are used to find a pattern from the UCI machine learning repository. The result is
inpatient health data. The best system is one that is the extracted with the base of accuracy. KNN produced 95%
efficient, adoptive, generic and affordable system. Many highest accuracy with 10 features [14].
factors affect the result of analysis like an error in online Sara Belarouci et al. propose meta-heuristics optimization
databases, sensor's settlement. This study shows that ASP logic methods for improvement of medical classifier performance.
approach is the best use for incomplete biological data. They are used many algorithms like Genetic Algorithm PSO,
Artificial Neural Network is best used for single purpose Simulated Anneeling to compare with Least Square Support
system. ANN generates best better result than ASP and another Vector Machine to improve the classification with aspect to
approach used in the health care system. If the hardware is False Positive and Negative. Meta-heuristics Optimization is
costly then it difficult to use this system [9]. best for solving the problem of unbalance dataset. Five
There are many data mining algorithm available but this different datasets related to various diseases like Liver
study provides a comparative study of three algorithms Naïve Disorder, Appendicitis, and Diabetes. This approach will help
Bayes, Decision Tree and Multi-Layer Perceptron Neural doctors to diagnose many diseases effectively [15].
Network. In this study, window operating system 8.1 is used Tharaha S and Rashika K proposed this research using
with WEKA data mining tool. Ebola Disease data set contain Hybrid Artificial Neural Network and Decision Tree algorithm
the range of 250-10000 instances that are stored in MySQL. for disease recognition. They used Artificial Neural Network
According to this study, the Naïve Bayes algorithm shows a for training data and decision tree for classification of data
negative correlation, with the increase in the dataset it because the Decision Tree algorithm is a good classifier.
performance lead to a decrease. WEKA shows a positive Datasets are taken from the human blood detecting and sensor
correlation. Naïve Bayes is the best and popular machine counting, stored with different attributes. Time taken for test
learning algorithm is fast in training [10]. split in ANN is 0.09s and where decision tree took time is
Mohammed H. Tafish and Dr. Alaa M. El-Halees proposed 0.14s. The result is shown by apply WEKA 3.8.1 version. The
a model as Breast Cancer Severity Degree Predication Using combination of these two algorithms gives the best result than
Data Mining Techniques in the Gaza Strip described that in separate used and provide the best help for disease diagnosing
Gaza Area cancer disease and diabetes growth are top disease [16].
during the last decades. They used a data mining method to Dania Abed aljawad et al. proposed an empirical study of
diagnose cancer and diabetes disease. They proposed a model Bayesian Network and Support Vector Machines for Breast
using data mining techniques like SVM, KNN, and ANN. Cancer surgery Survivability Prediction. They used
Breast cancer data taken from Gaza hospital used, after Haberman's survival dataset and evaluate the performance of
evaluation and test by applying the above techniques they the Bayesian network and Support Vector Machine using
obtain 77% accuracy for the prediction of the severity of breast WEKA tool. Empirical research shows that Support Vector
cancer [11]. Machine best performs with an accuracy of 74.44% than
Manickam Ramasamy at el. proposed a model for Bayesian network with an accuracy of 67.56%, Imbalance data
predicting hepatitis in which they provide an empirical analysis is converted into balance. This study helps the doctors to the
of the decision tree algorithm by using Hepatitis data set taken prediction of the patient stage of cancer using old data as a
from the UCI machine learning repository. They used different sample to new data [17].
310 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
P. Hamsagayathri and P. Sampath proposed a Priority destruct the predictive performance. There are two basic
Based decision Tree Classifier for Breast cancer. Women strategies for pruning the decision tree i.e. first forward pruning
mostly from 40-70 age affected with breast cancer. So they means pruning before completion of decision tree and other
proposed a model for prediction of breast cancer. Classification post-pruning means pruning after making a decision tree. So
provides a vital role in the detection of breast cancer and helps forward pruning stop the pruning process before reaching its
the researcher to analyze and classify data. SEER breast cancer maturity level and in a post-pruning button-up, approach is
data set is used in this paper. Two decision tree algorithm J48 used to cut off the node. The Minimum Description Length
and priority-based decision tree algorithm are used. The Principle, Expected Error Rate Minimization Principle and
priority-based algorithm provides the best result with less time Principle of Occam's Razor are used for pruning.
consuming to build the model. J48 used repetitive but priority
base algorithm not used repletion step and 98.51 accuracies A. ID3
[18]. ID3 stands for Iterative Dichotomize 3. It is built by J.R
Quinlan in [19]. It is the core algorithm to build a decision tree.
With the reference of above literature review, the specific It generates all possible decision tree. It simply classifies the
medical data sets i.e. arrhythmia, breast-cancer, diabetes, training and testing set for the dataset. It does not require much
hepatitis, mammography, lymph, liver-disorders, sick, more computation as compared to another approach for
cardiotocography, heart-statlog, breast-w and lung-cancer are creating a decision tree. It is an iterative approach. It chooses
not used to investigate the performance of Decision Tree (i.e. the training set randomly and makes the decision tree. If it
AdaBoost.NC, C45-C, CART, ID3-C) and Support Vector answers all object then it terminates the process it not then it
Machines. In this research study will Decision Tree based add to again in training data for further process. It iterates the
classifiers and SVM Machines for the discovery of process and makes the decision tree correctly up to thirty
classification rules. The problem statement and objectives of thousand instance and fifty attributes. This algorithm based on
this research are given in the next sections. the information gain of candidates attributes. If any attribute
III. DECISION TREE BASED CLASSIFICATION has more gain information then it selected for decision tree and
less gain information is discorded.
After Decision Tree is most popular supervised machine
learning algorithm applied for the various classification The effectiveness of this approach also depends on the
problems. It is used for classification and regression problems. computational requirement based on the gain of untested
Decision tree provides the result which is easily understandable attributes and non-leaf nodes of the decision tree. The total
by humankind. A decision Tree provide output in a tree-like computational power of the ID3 is relative to the size of the
graph in which each node represents to attribute, each branch training set, several attribute, and non-leaf nodes. The
provide a rule and each leaf node provide a target class. Target similarity in attributes extends the computational requirement.
class may be in discrete or in continuous form. Decision Rule In ID 3 time and space are not grow exponentially so it can be
may be in IF-then-Else rule. Big decision tree means the more used for larger and complex tasks.
complex rule. ID3 algorithm has some advantages like i.e. easily
Decision Tree is used as a top-down approach for making a understandable rule for classification, it is fastest and provides
decision tree. It begins from the root node to the leaf node. The a short tree. It calculation time is a linear function not
decision is made on each internal node where attributes are exponential as well as it has some disadvantages i.e. data may
split into further node if it contains information that can be be overfitted or over-classified due to the small sample and for
divided further. More information leads to further the continuous value it computation time may be more due to
classification. If a node cannot have information more then it make many trees to find where to break the continuum.
considered as leaf node that refers to the target value. B. C4.5
Different methods are used to construct a decision tree. Quinlan et al. proposed the extended version of ID3 that is
Every method used different information for the construction known as C4.5 in [20]. It is also developed for making a tree. It
of a decision tree. Large decision tree not considered an is developed by Quinlan in 1993. Quinlan described many
accurate and efficient decision tree. Different research shows issues for decision tree-like handling missing value, pruning
that the best decision tree is as small as possible. It based on and converting trees to rule and how C4.5 handle it. Decision
the proper selection of attributes. Attributes selection measures tree algorithms used some cases and make a tree-like structure
are used to split attributes into further sub attribute. It is a in which the main node is called the root node and other node
recursive approach. Attributes selection measure checks the are test node and leave node. Every decision node used a test
impurity of the attribute. Impurity measurement method and leave node show the class label.
includes Gain Ratio, distance measures, Gini-index and
information gain. ID3, C4.5 focused information gain and C4.5 algorithm creates a small, accurate and fast decision
CART use Gini-index for attributes selection. tree and it is known as a reliable classifier. These are the best
and popular properties for making the classification. This
A decision tree process can be divided into two steps: one algorithm extracts the best information from a set of cases and
constructs a decision tree and other to pruning a decision tree. takes only one attributes for the test. For this purpose
Data mining works on real world data. Data may have some information gain and gain, the ratio is used for the selection of
missing value, wrong value, containing noise or even less best attributes. Some dataset may contain unknown
essential data, so this problem may lead to over-fitting and will information so Quinlan used C4.5 approach. Information gain
311 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
for unknown value can be ignored. And known value attribute performance in analyzing subproblem as compared to combine
information gain can be calculated. So information on this test the whole problem. So it needs to better technique to combine
case may be quite small. The unknown value may affect the the subproblem to acquire knowledge from AdaBoost.NC.
decision tree making process.
D. CART
An every decision tree cannot be considered as a good CART stands for classification and regression tree. CART
classifier for every data set in respect of making a smaller tree is proposed by Breiman et al. in [22]. It is an algorithm used to
that may not fit for all training data. So avoid by overfitting, construct a decision tree from the categorical and continuous
many decision tree algorithm used the pruning method. In this form of data. Classification is used for a categorical form of
method, growing the decision is stopped while deleting the data and regression tree is constructed from a continuous form
portions of the tree. C4.5 pruning method based on error rate. of data. The first time Morgan and Sonquist proposed a method
The error rate of every subtree is calculated if the error rate is to construct a tree by quantitative variable. They gave the name
low then it will be treated as a leaf node. This process used Automatic Interaction Detection. Each cluster is grouped into
bottom-up approach. If C4.5 algorithm indicates that tree will two clusters. Each predictor is tested on every cluster. Their
be treated as accurate even children of concern node deleted model naturally incorporates interaction among all predictor.
than algorithm considered concern node as a leaf node. If this
method proved as good then this decision tree is considered the A classification tree is dependent on discrete or categorical
best decision tree. value. Kass (1980) proposed a modification in AID model
called CHAID for the creation of a tree from the dependent and
Quinlan discusses some shortcoming of c4.5. It has a built- independent variable. This model limited to categorical
in bias, t take only a single attribute for testing that takes more predictor so it cannot be used for the quantitative variable.
time computation. It makes the value of the given attribute in
the same group and considered as a single value. It may use for These two models have a problem where to stop the tree.
single training set once and not used for other training set for Breiman et al. (1984) method show that node that cannot
binary classification. Suppose attributes for a chemical element contribute to prediction eliminate from the tree.
that can be classified into the light and heavy element and other CART is a mechanism to construct a decision tree. It makes
training set having an electric conductor that can be classified the solution in a tree-like structure. It starts from the root node
in conductor and non-conductor. So these groups may overlap and split into a test node on the base of selected attributes. This
with each other. This algorithm cannot is used for both groups. process ends on the leaf node that cannot be further divided. To
C4.5 used greedy approach for the grouping, so it gives the make the best and effective tree it used pruning method i.e.
unsatisfactory result and remains an open problem. Complexity based pruning. Pruning is started from the bottom
C. Adaboost.NC toward the root node.
AdaBoost.NC is a negative correlation learning algorithm CART algorithm may a structure of question and answer of
proposed by Wang et al. in [21]. It is used for classification these question lead to the next question. So, the result of these
ensemble. AdaBoost.NC algorithm is used for multiclass question make a tree structure where to question is not more.
imbalance data. It provides the solution of two class imbalance CART uses the basic rule for making a decision tree i.e.
problem. AdaBoost.NC provides the best accuracy with splitting data rule and stopping rule where the terminal cannot
random oversampling on the minority class as compared to be split and prediction of the leaf node. CART has some
another balancing approach. The accuracy is achieved by the advantages like can handle missing value automatically.
less border classification and overfitting in the minority class.
IV. SUPPORT VECTOR MACHINE BASED CLASSIFICATION
AdaBoost.NC is the advance version of AdaBoost for
negative correlation but it based on AdaBoost training Support Vector Machine was introduced after in the 1990s
framework. It provides better classification boundaries and and used for many engineering application [23]. Support
creates lower error correlation as compared to AdaBoost. This Vector Machine is an algorithm developed for binary
is used to improve the performance of the original AdaBoost classification by Cortes & Vapnik. The objective of this
algorithm. This algorithm is used for better classification in algorithm to find hyper-plane and classification of data points.
control of upper bound on the generalization error of It is used for separating the two classes with a maximum
Traditional AdaBoost. AdaBoost.NC provided the best margin between two points called support vector. SVM
performance in respect of the distribution of better margin. algorithm is used for class separation, nonlinearity and
overlapping classes where a data point lies in the opponent
AdaBoost is a very simple and effective ensemble class [24].
algorithm. It is not only used to emphasize to misclassified
example, but also provide the mechanism to control the error of Support Vector Machine classifies the data by using hyper-
misclassification of the same example. Due to this reason, it plane. The hyper-plane can be chosen by either of the sides but
provides the best accuracy and diversity. optimal hyper-plane is that maximizes the margin between two
support vectors. Support vector is the data point that closer to
AdaBoost.NC does not show good performance in overall the hyper-plane. Hyper-plane has different features on different
and in minority class working with class decomposition location and deleting the support vector can influence the
scheme. This algorithm receives and learns from all data position of the hyper-plane [25].
information of all classes. It learns from several decomposition
problems for partial knowledge. It provides the best
312 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
313 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Data Sets Name No. of Attributes Attributes Type Missing Value No. of instance
Lung-cancer 56 Integer 2 32
B. KEEL paper and final module provide the results of classifiers for the
specific databases.
Knowledge Extraction based on Evolutionary Learning
(KEEL) is a data mining tool possessing various facilities for VI. RESULT AND DISCUSSION
data preprocessing and different types of classification
approaches for the comparison of new proposed classification This section provides the performance analysis of decision
approaches. It is a freeware java software tool. It provides a tree based classification approaches and support vector
user-friendly GUI interface. It contains many built-in dataset machines on medical databases in terms of accuracy and
and algorithm for data analysis. It provides many preprocessing variance. Furthermore, the performance of a specific classifier
techniques like feature selection, a method for missing value is investigated in two fold; on a specific medical database and
and hybrid models and statistical method for experiment.it use among the classification approaches.
for educational and research purposes [8]. A. Performance Analysis of Decision Tree based Classifiers
The current version of Keel has many advance features like Table II shows the comparative performance analysis of
multi-instance learning, subgroup discovery, semi-supervised AdaBoost.NC-C C4.5 –C, C4.5_Binarization–C and CART-C
learning and imbalanced classification. These features make tree based classifiers that are chosen in this empirical research
versatility of the Keel improved and better deal with new data study. We compare the performance of these algorithms in
mining problems [29]. Table II on different datasets in term of accuracy. The results
show that C45-C and C45_Binarization-C provide equal
C. Experimental Graph accuracy on lung-cancer dataset. Moreover, C45-C also
Fig. 1 shows the experimental graph generated in the perform better on lymph, primary-tumor breast cancer dataset
KEEL. First stage data set loading, the second stage provide as compared to other algorithms in terms of accuracy.
the facility of the imputation of missing values, the third stage C45_Binarization provide the best performance in term of
provides the module for data discretization, the fourth stage accuracy on Dermatology and Heart-statlog dataset. The
shows the algorithms exploited the empirical study in this AdaBoost.NC-C provide promising results on Hepatitis
314 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
dataset; CART-C provides the best performance on Haberman More variance on dataset provides lower performance result.
dataset while the C45-C classifier provides 75.19% average CART-C provide bad performance on Lung-Cancer as well as
accuracy on all datasets that is more promising comparatively on Dermatology and Hepatitis datasets as compared to
w.r.t other classification algorithms. The C45-C_Binerization AdaBoost.NC-C, C45_C, and C45_Binarization. C45-C
provide minimum accuracy of 6.06% and maximum accuracy provide variance on two selected dataset such as lymph and
96.05 in percentage. Table III shows the comparative primary tumor and AdaBoost. NC-C classifier provides more
performance of the selected classifiers in terms of variance on Breast cancer, Haberman, and Heart-Statlog.
win/lose/draw. The win/lose/draw provides information, how C45_Binarization –C classifier provides the best performance
many times a specific algorithm best performs to others. on selected dataset because there is no more variation as
compared to other proposed classifier. CART-C provide 1.01%
From Table III, C45-C provides best accuracy on 4 selected average variance and maximum 4.27% variance on selected
datasets with respect to other classifiers. AdaBoost.NC and dataset. C45-C provide minimum variance of 0.13% that is
CART-C provide best accuracy only on one dataset and more than the other three classifiers. Fig. 2 provides more
remaining 7 dataset loose by others algorithm. So understandability of this decision tree based classifier's
AdaBoost.NC and C45_Binarization draw in one dataset. variance.
The application of decision tree based classifier on selected
dataset also provides performance in term of variance parallel.
TABLE. II. DECISION TREE BASED CLASSIFIERS PERFORMANCE IN TERM OF ACCURACY (%)
4.50
4.00
3.50
Variacne (%)
3.00
2.50
2.00
1.50
1.00
0.50
0.00
Data Sets
315 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 3 provides the comparative performance analysis of Average 74.36 65.84 73.05
SVMs in terms of variance. SMO-C provide more variance on
primary-tumor, Dermatology and hepatitis than NU_SVM-C Min 44.76 33.80 46.12
and C_SVM-C and also provide minimum variance of selected
Max 95.79 97.03 97.28
variance as compared to other two classifiers. NU_SVM-C
provide more variance on six datasets that make the
performance bad on selected dataset as compared to other TABLE. V. COMPARISION IN TERMS OF WIN/LOSE/DRAW
datasets. It also make more value of average variance on Support Vector Machine Based Classifiers
selected datasets that reach 1.26. NU_SVM-C and C_SVM-C
provide equal maximum variance on selected datasets; as well SMO NU_SVM C_SVM
as equal variance on lung-cancer dataset. All the information is Win 6 0 4
highlighted in Table V.
Loose 3 7 3
Table VI provides the combined performance behavior of
both categories Decision Tree-based classifiers and SVMs Draw 0 1 1
based classifiers in terms of accuracy. The performance of
316 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
3.50
3.00
2.50
Variiance (%)
2.00
1.50
1.00
0.50
0.00
Data Sets
VII. CONCLUSION performs better in terms of a total number of datasets while the
overall average performance of C45_Binarization-C is better
Classification Rule Discovery from medical databases is a than other decision tree-based classifiers. The performance of
very hot and challenging problem in the field of Data Mining. SVM based classifiers, SMO-C is results are promising to
There are several classification approaches proposed for the NU_SVM-C and C_SVM-C in terms of accuracy. This
discovery of classification rules and prediction of diseases from research work provides the empirical performance analysis of
medical databases. The choice of a classification method for decision tree-based classifiers and SVM on a specific dataset.
the discovery of classification rules from specific medical Moreover, this paper provides a comparative performance
databases still requires investigation of the suitability of analysis of classification approaches in terms of statistics.
classifiers in terms of performance analysis. This study
investigates the performance of decision tree-based classifiers In the future, this research work can be enhanced by
and Support Vector Machines on specific medical databases. increasing the number of medical databases with other
The empirical performance analysis results reveal that C45-C statistical and evolutionary classifiers.
317 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
318 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Modern medical practice has embraced facial filler easily identifying whether there are similar cases that have
injections as part of the innumerable cosmetic procedures that been registered before, and then returns the diagnosis and
characterize the current age of medicine. This study proposed a treatment of these similar cases [1]. There is still a dearth in
novel methodological framework. The Inception model is the research studies regarding how computer vision can be
core of the framework. By carefully detecting the classification of deployed in facial filler injections.
wrinkles, the model can be built for different applications to aid
in the detection of wrinkles that can objectively help in deciding This study presents a novel methodological framework.
if the forehead area needs to have filler injections. The model The Inception model is the core of the framework. With
achieved an accuracy of 85.3%. To build the Inception model, a carefully detecting the classification of wrinkles, the model can
database has been prepared containing face forehead images, be built for different application to aid the detection of wrinkles
including both wrinkled and non-wrinkled face foreheads. The to that can objectively help in determining if the forehead area
face image pre-processing is the first step of the proposed needs to have filler injections.
framework, which is important for reliable feature extraction.
First, in order to detect the face and facial landmarks in the II. RELATED WORK
image, a Multi-task Cascaded Convolutional Networks model has
Previous studies have also proposed a new algorithm
been used. Before feeding the images into the deep learning
Inception model for classifying whether the face foreheads have
referred to as Hessian Line Tracking (HLT) for detecting
wrinkles or no wrinkles, an image cropping process is required. wrinkles [2]. The researchers began with a group of seeds that
Given the bounding box and the facial landmarks, face foreheads the researchers extracted from the Hessian Matrix‟s ridge area
can be cropped accurately. The last step of the proposed before proceeding to apply a multi-scale tracking system
methodology is to retrain an Inception model for the new recursively to all the seeds. After completion, the researchers
categories (Wrinkles, No Wrinkles) to predict whether a face validated each pixel confidence over the scales with the
forehead has wrinkles or not. objective of producing an initial map of wrinkles. The last step
involved post-processing in which the researchers carried out a
Keywords—Deep learning; classification; facial fillers; wrinkle series of mini-steps including median and directional filtering
detection as well as area thresholding in order to reduce noise [2]. In the
experimental set-up, the researchers employed the services of
I. INTRODUCTION three number of coders instructed to manually annotate the
Filler injections are a modern cosmetic procedure and have wrinkle on 100 cropped images of the forehead– the images
been widely embraced by women and men alike because of were extracted from the Bosphorus dataset [3]. The dataset
their wonderful ability to create fuller cheeks, lips, and other contains forehead wrinkles of varying sizes. Particularly, the
facial features. Filler injections are also used to reduce the dataset that was employed in the study contained 106 subjects
effects of wrinkles around the mouth, eyes, and eyebrows and from whom the researchers used an ordinary camera to fetch 2-
to hide any scars that may be causing an individual to feel self- D facial images under good flat, illuminated conditions.
conscious and unattractive. Although the researchers took several images of varying poses
and facial expressions of each subject, the experiment only
Human beings can without much difficulty see what the used frontal images. The researchers reported both intra and
image represents. As an example, humans can easily see that inter-reliability with regard to the manual annotation process –
the image contains a number of objects and can detect faces in reliability was 94% and above [2]. Wrinkles typically appear in
an image as well as distinguish between the different features a wide variation in both images, in pattern, length, and width as
of the face. Computer Systems, on the other hand, can have well as within the same image. This significantly challenges
difficulties. Computers cannot easily see whether the image the generation of an automatic wrinkle detection operator. This
contains objects or not. Also, cannot easily detect human faces justified the development of a multiscale HLT – an approach
and facial features. based on seed extraction by Hybrid Hessian Filter (HHF) as
As many researchers are trying to set up computers with well as multiscale tracking for overcoming the weakness of
artificial intelligence capabilities to be able to serve patients HHF while also making it possible to capture wrinkle
with different health problems such as diabetes, blood pressure, variability in the entire image.
and obesity [1]. Artificial intelligence databases can help in
319 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In another study researchers also proposed a new algorithm conclude is the most ideal tool for in-painting irregularly
for automatic tracking of linear, fixed and chaotic forms of shaped gaps that are left behind in the form of scars of
transient wrinkles [4]. For the automatic analysis of wrinkles, removed wrinkles [6]. Overall, the experiment illustrates that
the researchers came up with two clusters of wrinkles including most skin imperfections and wrinkles are usually detected and
transient and permanent. While the latter are usually found on in-painted. Nevertheless, there are a few areas on the face that
the faces of older individuals, the former on the other hand have less contrast to the skin around it that fails to be detected.
often appear in relatively wider regions in the course of A case in point includes small parts of the upper parts of an
generating an expression. The research was divided into two individual‟s forehead. Moreover, the experiment also reported
parts whereby while the first part explored an algorithm for challenges of the impact of sagging skin and aging,
detecting transient wrinkles, the second part dwelled on its illumination, as well as artefacts resulting from the repetition of
application. The proposed wrinkle detector was made up of patches. Particularly, facial images of subjects with sagging
three steps. The first step involved the Canny edge detector that skin that appears alongside wrinkles according to the
the authors applied to the input face for detecting pairs of researchers posed significant challenges. Such is the case since
continuous wrinkles. This was followed by applying an active the same patch of wrinkled skin is selected as a source of skin
appearance model to locate all candidate wrinkle lines. This texture resulting in repeated patterns of imperfection.
generated data for constructing the structure of the wrinkle. In
the third step, the researchers defined quantitative metrics III. BACKGROUND AND METHODOLOGY
which they subsequently used for Support Vector Machine A. Deep Learning
(SVM) classification – this step was critical in helping to
discriminate regions of the face with wrinkles from those Deep learning is one of the major theories of machine
without. Despite the fact that competitive results are achieved learning that is founded on learning data representation and not
by the proposed transient wrinkle detector model that the task-related algorithms. It was first introduced in 1986 to the
researchers propose which is also the case with improved community of machine learning by Rina Dechter [7]. Learning
wrinkle mapping, there is a number of areas in which future using this theory might be unsupervised, supervised or a hybrid
studies should focus. A case in point regards long wrinkles in of both. As a branch of machine learning, deep learning is
the forehead, for which [4] employed the five points in the inspired by the human brain‟s main function and structure. The
wrinkle structure–future studies should add on more points brain constitutes of neural networks whose interconnected
which will go a long way towards improving the neurons plays a crucial role in processing and transmitting
approximation of accuracy of the wrinkle edge. signals from one neuron to the other. On the basis of this
operation, the founder of deep learning, Geoffrey Hinton [8],
Batool and Chellapa [5], in their work presented a quick made some artificial neural networks comprising of man-made
deterministic algorithm based on image morphology as well as neurons that could easily conduct operation as well as process
Gabor filters with the aim of improving localization results. the required information. The three layers of the neural
The researchers proposed features derived from Gabor filter network in deep learning include the input, hidden and output
bank–the aim here was to shed light on the subtle curvilinear layers. The input layer is responsible for accepting a variety of
discontinuities in the texture of the skin attributed to wrinkles. input using formats such as audio, picture, number or text. The
The researchers then employed image morphology for hidden layer conducts mathematical functions, feature
integrating geometric constraints to localize curvilinear wrinkle extraction and manipulation of data. The output layer, on the
shapes at the locations of wrinkles of pronounced Gabor filter other hand, is essential in getting the desired final output [8].
responses. Experiments were carried out at two sets of images
including those with high and low resolutions before the B. Transfer Learning
researchers compared the results to those generated from In 1993, Lorien Pratt came up with algorithm that was
Marked Point Processes (MPP). Experiments illustrate that the founded on discriminability-based transfer [9]. In so doing, he
suggested algorithm is not only faster compared to the MPP gave a platform through which the transfer learning theory was
framework, but also generates the merit of visually satisfactory born. In this method, a model that has been formulated for an
results. activity is normally reutilized as a beginning point for some
other second task. For instance, the knowledge that one gets
In the last study analysed in the review, researchers when learning to distinguish a jet could be useful in
compounded texture orientation fields with Gabor filter recognizing a helicopter or a spacecraft. Therefore, this model
responses to detect wrinkles [6]. In the experiment, a bimodal concentrates on the storage of acquired knowledge while at the
Gaussian Mixture Model (GMM) described the distribution of same time striving to be utilized in another similar scenario or
normal skin verse skin imperfection of Gabor features [6]. The problem [10]. This theory is used in deep learning particularly
researchers then proceeded to employ a Markov random field in instances where the pre-trained models are employed as
model to integrate the spatial relationships for their texture beginning points on the natural language and computer vision.
orientations between adjacent pixels as well as GMM In comparisons to the conventional machine learning, transfer
distribution. To classify skin versus skin imperfections, the learning utilizes these pre-trained models that were useful in
study employed an expectation-maximization algorithm. As another instance to kick start the process of development of the
opposed to blending or blurring the detected wrinkles, the new problem or task. As stated earlier, this model depends on
study removed them completely. The exemplar-based the task and domain concepts [10].
constrained texture synthesis algorithm, as the researchers
320 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
C. Multi-Task Cascaded Convolutional Neural Network using a Multi-task Cascaded Convolutional Networks model
Multi-Task Cascaded Convolutional Neural Network [11], known for its strong and accurate abilities to detect faces
model or simply MTCCNN is a deep learning algorithm-based and facial landmarks quickly. Before feeding the images into
model that comprises of three stages that identify faces‟ and the deep learning Inception model for classifying whether the
bounding box in a particular image as well as the main five- face foreheads have wrinkles or no wrinkles, an image
point landmarks on an individual‟s face [11]. Back in 2001, cropping process is required. Given the bounding box and the
researchers attempted to put forward a method of forward- facial landmarks, face foreheads can be cropped accurately.
cascade detection founded on the features of AdaBoost and Next, k-means algorithm has been used to separate 1,199-
Haar to conduct cascade classifiers [12]. Kaipeng Zhang later image data set of face foreheads into two clusters; then went
introduced the MTCNN theory which in addition to having through each cluster and separated wrinkled foreheads from
three phases, helped in the detection of the bounding box, non-wrinkled foreheads. Also removed the foreheads covered
points, and landmarks [11]. by hair (581 images), then labelled images Wrinkles (309
images) and No-Wrinkles (309 images). The last step of the
For every MTCNN phase that the image goes through, the proposed methodology is to retrain an Inception model for the
investigator sees an improvement in quality. The input goes to new categories (Wrinkles, No Wrinkles) to predict whether a
the CNN which not only give it a particular score but also face forehead has wrinkles or not.
return a bounding box. The initial stage ensures that the input is
scaled downwards. The CNN also ensures the facilitation of B. Dataset and Pre-Processing Details
the MTCNN in making a pyramid of a picture in question. The The FERET database [14], [15] is used in the
next stages involve the extraction of the patches of this picture implementation experiments to evaluate the performance of the
for every bounding box. It is then resized and resized even proposed methodology. The FERET database was established
more in the third stage. Other than bounding the image onto the to support machine learning algorithms in both development
box and later assigning some score, this stage also computes and evaluation. The database contains 14,126 images
the points of the five face landmarks for every bounding box associated with 1,199 people. In the experiments only the
[11]. frontal face images have been used from the database, which
amounted to 1,199 unlabelled images.
D. Inception-v3 Model
Inception V3 is one of the most popular image-recognition To detect faces in an image and discover the location of
deep learning models. It has been a culmination of multiple different facial features a pre-trained Multi-task Cascaded
ideas by a pool of researchers in the years. Inception V3 was Convolutional Networks has been used [11]. Face alignment
theorized by Szegedy in a paper that rethought the Inception and face detection are executed jointly in a multi-task training
concept in computer vision [13]. In itself, this model is method, enabling the model to properly detect faces and locate
composed of both asymmetric and symmetric building blocks. five points of facial landmarks.
They include convolutions, max, and average pooling as well By default, the pretrained Multi-task Cascaded
as fully linked layers. Throughout the Inception –v3 model, Convolutional Networks packaged with a face detection
Batchnorm is extensively utilized in addition to being applied weights model. The detector passes a list of Javascript object
to the activation inputs. On the other hand, Softmax is usually notation objects. Each JavaScript object notation object carries
used in computing the loss. The two parts of Inception-v3 three main keys: 'confidence', 'keypoints', and „box':
model include the feature extraction component and the
classification component. The former components are reliant The confidence is how probable it is that a bounding
on the convolutional neural networks while the latter is reliant box will match a face.
on softmax and fully-linked layers [13]. The box is put into a format [x, y, W, H] which can
E. Dataset produce the bounding box around the face.
The FERET database [14], [15] is used in this study to The keypoints are arranged into a Javascript object
experiment and evaluate the performance of the proposed notation object with the keys 'eyeLeft', 'eyeRight',
methodology. The FERET database was established to support 'nose', 'leftMouth', and 'rightMouth'. Every keypoint is
machine learning algorithms in both development and recognized by the position of the pixel (x, y).
evaluation. The database contains 14,126 images associated
with 1,199 people. Fig. 2 shows a result of an image with a bounding box
around the face and five landmarks located on the eyes, mouth,
IV. EXPERIMENTS AND RESULTS and nose after it has been detected by the MTCNN model. An
image cropping process is required. Given the bounding box
A. Overview of the Proposed Methodology Framework and the facial landmarks, face foreheads can be cropped
As shown in Fig. 1, the core of the framework is the accurately.
Inception model. To build the Inception model, a database has
been prepared containing 618 cleaned face forehead images, As regards the labelling process and cleaning data, the
including both wrinkled and non-wrinkled face foreheads. The manual labelling of images would be a hurdle, so the k-means
face image pre-processing is the first step of the proposed unsupervised learning clustering algorithm has been used to
framework, which is important for reliable feature extraction. help with this process.
First, the face and facial landmarks are detected in the image
321 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
322 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
323 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 5. The Learning Curves for the Training and the Validation of Inception Model.
In order to ensure that overfitting is not occurring in the re- Since there is a lack in the previous studies there is much to
trained Inception model with the new dataset and labels, as do in the future research studies. For the presented research,
shown in Fig. 5, a line plot that displays the learning curves for there is only one part has been investigated so the other area of
the training and the validation of the Inception model. The the face to be investigated such wrinkles around the eyes and
curve for the training set is slightly more accurate than the test mouth.
set to avoid overfitting.
Semantic segmentation methods can be used to label each
V. DISCUSSION pixel of the wrinkles, and fine lines in the face so that the
person can know exactly the location of the areas that need to
One of the main limitations of this study that have been be filled with filer injection instead of classifying whether an
encountered during the investigations of the proposed research area of the face has wrinkles or not. For more complicated
is the lack number of previous studies that explore how applications it could also be used two frontal face pictures of
computer vision methods can help in the facial cosmetic the same person, one image with a blank expression and the
healthcare field. other image with happy facial expression. The model will then
Furthermore, the resolution quality of FERET images compare those two images against each other and then decide
database used in this study is low compared to today‟s image if the wrinkles show only in the happy facial expression then
resolutions also the time constraint allowed only to study one it‟s transient wrinkles and have to be filled with Botox, if the
area of the face which is forehead. wrinkles appear in the blank facial expression then it needs to
be filled with natural substances such as body fat or collagen
VI. CONCLUSION AND FUTURE WORK and artificial substances such as hyaluronic acid. Essentially
Experiments accomplished in the study investigate a that would be like an artificial intelligent cosmetic consultant.
proposed novel methodological framework. The Inception
ACKNOWLEDGMENT
model is the core of the framework. By carefully detecting the
classification of wrinkles, the model can be built for different “Portions of the research in this paper use the FERET
applications to aid in the detection of wrinkles that can database of facial images collected under the FERET program,
objectively help in deciding if the forehead area needs to have sponsored by the DOD Counterdrug Technology Development
filler injections. The model achieved an accuracy of 85.3%. To Program Office” [14], [15].
build the Inception model, a database has been prepared REFERENCES
containing face forehead images, including both wrinkled and [1] P. Hamet and J. Tremblay, "Artificial intelligence in medicine",
non-wrinkled face foreheads. The face image pre-processing is Metabolism, vol. 69, pp. S36-S40, 2017. Available:
the first step of the proposed framework, which is important for 10.1016/j.metabol.2017.01.011.
reliable feature extraction. First, the face and facial landmarks [2] C. Ng, M. Yap, N. Costen and B. Li, "Wrinkle Detection Using Hessian
are detected in the image using a Multi-task Cascaded Line Tracking", IEEE Access, vol. 3, pp. 1079-1088, 2015. Available:
Convolutional Networks model known for its strong and 10.1109/access.2015.2455871.
accurate abilities to detect faces and facial landmarks quickly. [3] A. Savran, B. Sankur and M. Taha Bilge, "Regression-based intensity
estimation of facial action units", Image and Vision Computing, vol. 30,
Before feeding the images into the deep learning Inception no. 10, pp. 774-784, 2012. Available: 10.1016/j.imavis.2011.11.008.
model for classifying whether the face foreheads have wrinkles [4] W. Xie, L. Shen and J. Jiang, "A Novel Transient Wrinkle Detection
or no wrinkles, an image cropping process is required. Given Algorithm and Its Application for Expression Synthesis", IEEE
the bounding box and the facial landmarks, face foreheads can Transactions on Multimedia, vol. 19, no. 2, pp. 279-292, 2017.
be cropped accurately. The last step of the proposed Available: 10.1109/tmm.2016.2614429.
methodology is to retrain an Inception model for the new [5] N. Batool and R. Chellappa, "Fast detection of facial wrinkles based on
categories (Wrinkles, No Wrinkles) to predict whether a face Gabor features using image morphology and geometric constraints",
forehead has wrinkles or not.
324 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Pattern Recognition, vol. 48, no. 3, pp. 642-658, 2015. Available: [11] K. Zhang, Z. Zhang, Z. Li and Y. Qiao, "Joint Face Detection and
10.1016/j.patcog.2014.08.003. Alignment Using Multitask Cascaded Convolutional Networks", IEEE
[6] N. Batool and R. Chellappa, "Detection and Inpainting of Facial Signal Processing Letters, vol. 23, no. 10, pp. 1499-1503, 2016.
Wrinkles Using Texture Orientation Fields and Markov Random Field Available: 10.1109/lsp.2016.2603342.
Modeling", IEEE Transactions on Image Processing, vol. 23, no. 9, pp. [12] P. Viola and M. Jones, "Rapid object detection using a boosted cascade
3773-3788, 2014. Available: 10.1109/tip.2014.2332401. of simple features", Proceedings of the 2001 IEEE Computer Society
[7] R. Dechter, "Learning while searching in constraint-satisfaction- Conference on Computer Vision and Pattern Recognition. CVPR 2001,
problems", Proceeding AAAI'86 Proceedings of the Fifth AAAI 2001. Available: 10.1109/cvpr.2001.990517.
National Conference on Artificial Intelligence, pp. 178-183, 1986. [13] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens and Z. Wojna,
[8] A. Krizhevsky, I. Sutskever and G. Hinton, "ImageNet classification "Rethinking the Inception Architecture for Computer Vision", 2016
with deep convolutional neural networks", Communications of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR),
ACM, vol. 60, no. 6, pp. 84-90, 2012. Available: 10.1145/3065386. 2016. Available: 10.1109/cvpr.2016.308.
[9] L. Y. Pratt, "Discriminability-based transfer between neural networks", [14] P. Phillips, H. Wechsler, J. Huang and P. Rauss, "The FERET database
Proc. Adv. Neural Inf. Process. Syst., pp. 204-211, 1993. and evaluation procedure for face-recognition algorithms", Image and
Vision Computing, vol. 16, no. 5, pp. 295-306, 1998. Available:
[10] K. Weiss, T. Khoshgoftaar and D. Wang, "A survey of transfer 10.1016/s0262-8856(97)00070-x.
learning", Journal of Big Data, vol. 3, no. 1, 2016. Available:
10.1186/s40537-016-0043-6. [15] D. MacKay, Information theory, inference, and learning algorithms.
Cambridge: Cambridge University Press, 2003, pp. 284–292.
325 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
326 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
327 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
been obtained. So, all cardiac patterns are of dimension 301, TABLE. I. CLASSIFICATION RATE % VS. SPACE DIMENSION FOR LLP, LE
AND CS (SIGMA = 5, NEIGHBORHOOD K=9)
with the R wave placed on the 150-th sample.
Space dimension LE LPP CS
We constructed a database, namely a data set with 5608
2 82,61 48,10 40,64
patterns, having 701 patterns for each of the above classes. All
patterns were normalized to unity norms. However, sensitivity 3 87,46 72,55 71,47
to normalization was observed only in the case of LPP. 4 89,40 79,87 76,79
5 90,85 84,14 83,06
To classification in the initial 301 dimensional signal space 6 91,01 86,83 87,99
we have used the KNN classifier using the Euclidean distance
7 90,82 88,11 86,72
and the membership decision was based on the nearest
8 90,98 90,41 89,29
neighbor.
9 91,35 90,96 91,07
For the original normalized ECG data the classification rate 10 91,57 91,54 91,65
for the eight classes analyzed has been found to be 94.92% 12 91,48 92,76 92,12
[11]. 14 91,82 93,01 93,20
Fig. 1 shows the results for all tested methods for 16 92,07 93,21 92,32
dimensionality reduction. It can be seen that for LE and CS the 18 91,87 93,76 93,81
classification rate is increasing and once a maximum is 20 91,68 93,87 93,54
reached, the classification rate stabilizes around that value. For 22 92,07 94,15 93,73
very small values of the space dimension the best results are 24 92,15 94,34 93,90
obtained with LE. Thus, for space dimension equal to 2 a 26 92,23 94,29 93,78
classification rate of 82.61% is obtained, while for space 28 92,48 94,29 94,09
dimension 8 the results are comparable for all three tested
30 92,59 94,37 94,23
methods. If we refer to the maximum values achieved in terms
32 92,62 94,21 93,70
of classification, then LPP offers the best results, i.e., for space
dimension equal to 25, a 94.37% classification rate was 34 92,65 94,21 93,92
obtained [10]. Several results regarding dimensionality 36 92,65 94,26 93,76
reduction for the three tested methods are presented in Table I. 38 92,73 94,18 93,92
40 92,68 94,18 94,20
Since many techniques are noise-sensitive, we tested all 42 92,79 94,21 94,01
three algorithms with waveforms with 8% added noise
44 92,76 94,23 94,03
normally distributed.
46 92,68 94,23 94,39
Fig. 2 shows the classification rate obtained by LPP for 48 92,62 94,23 94,37
ECG segments with and without noise. It is found that the 50 92,65 94,18 93,90
locality-preserving character of the LPP method makes it 75 92,84 93,96 94,64
relatively insensitive to noise because the classification rate 100 92,95 93,87 95,00
varies significantly in the presence of 8% noise.
125 92,93 93,79 94,42
Fig. 3 shows the classification rate obtained by Laplacian 150 92,90 93,68 94,62
Eigenmaps for ECG segments with and without noise. There 175 93,17 93,68 94,92
are some small differences, but they are not significant so that 200 93,09 93,43 94,62
LE can be considered almost insensitive to the presence of
noise.
328 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
For dimensions less than 10, it has been found that the
classification rate for CS may differ, depending on the
projection matrix. For this we tested several projection
matrices; Table II presents these results and the average
classification rates. For all three methods, the sensitivity of the
algorithm to data normalization was analyzed as well.
Fig. 4 shows the classification rates obtained by
compressed sensed ECG patterns with and without noise. The
results are similar, so CS is not noise-sensitive as well.
Fig. 5 shows ECG patterns transformed into a 3-
dimensional space for LE (87.46% classification rate), LPP
(72.55% classification rate) and CS techniques (66.5%
classification rate).
Fig. 6 shows the classification rate obtained by LPP for
ECG segments with various levels of noise and without noise
for non-normalized signals. Interestingly, for dimensions less
than 40, the classification rates are practically unaffected by
noise. However, for larger space dimensions, the results are
worse but are improved by noise. Observe that, if we refer to
the maximum values achieved in terms of classification, then
LPP offers the best results for space dimension equal to 27
with a classification rate of 94%, even higher than that
obtained for the initial non-normalized ECG signals (space
Fig. 5. ECG Patterns Reprezented into a 3-Dimensional Space with LE, LPP
dimension equal to 301 and classification rate 92.5%). and CS Techniques.
In Table III, several results for the LPP method applied to
noiseless and noisy ECG waveforms are presented.
Fig. 4. Classification Rate % vs. Space Dimension with CS Noisy and Fig. 6. Classification Rate % vs. Space Dimension with LPP (Sigma = 5,
Noiseless ECG Normalised Patterns. Neighborhood k = 9) for Original and Noisy Non-Normalised ECG Patterns.
329 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. CLASSIFICATION RATE VS. SPACE DIMENSION FOR COMPRESSED SENSED WITH A FEW PROJECTION MATRICES
DIM mean CS 1 CS 2 CS 3 CS 4 CS 5 CS 6 CS 7 CS 8 CS 9 CS 10
2 39,09 44,64 33,19 36,56 37,50 40,33 40,75 38,47 36,53 42,25 40,69
3 53,22 53,31 54,94 50,42 54,81 49,72 48,44 61,31 48,03 64,39 46,86
4 66,03 66,19 63,50 66,81 68,83 62,50 68,83 63,78 74,22 62,06 63,56
5 73,20 74,53 80,19 74,31 71,97 70,08 69,75 74,56 68,00 72,50 76,11
7 79,63 83,47 81,17 74,28 78,89 77,94 83,58 80,44 77,92 76,33 82,22
9 84,54 84,89 84,31 84,17 84,25 85,31 84,92 85,03 84,78 82,42 85,31
11 85,68 85,56 82,44 85,47 84,94 86,28 87,25 87,42 87,28 85,94 84,19
13 86,86 85,11 88,39 89,03 85,72 85,44 87,75 85,56 86,97 88,11 86,50
15 87,98 87,94 87,06 87,56 88,36 88,00 87,31 88,28 87,83 88,94 88,53
17 88,21 89,50 88,56 89,33 87,78 87,67 87,67 88,22 87,69 88,42 87,22
20 88,99 89,75 89,00 90,03 88,72 88,75 89,92 88,86 88,14 88,75 87,94
25 89,12 89,72 87,94 89,00 87,94 89,14 89,94 89,81 90,42 88,72 88,58
30 89,33 89,69 88,83 89,31 88,92 89,00 88,75 89,86 90,50 89,00 89,42
35 89,95 90,14 89,86 89,36 89,64 90,56 90,19 90,56 89,81 89,19 90,19
40 89,76 90,14 89,75 89,44 89,11 90,83 89,25 89,17 90,17 89,78 89,94
45 89,81 89,83 89,50 89,94 90,17 89,14 90,19 89,97 89,44 89,72 90,17
50 90,14 90,50 89,31 89,94 90,22 89,86 90,64 89,89 90,25 90,83 90,00
TABLE. III. CLASSIFICATION RATE VS. SPACE DIMENSION FOR LPP WITH SEVERAL TYPES OF NOISE (FOR NON-NORMALISED SIGNALS)
Red color yellow color black color green color blue color dash-dotted red line
330 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Last, but not least, we found that of all three tested [4] M. Belkin, P. Niyogi, Laplacian eigenmaps for dimensionality reduction
algorithms, LPP is the most robust to noise but sensitive to data and data representation. Neural Computation 15:1373–1396, 2003
normalization, while CS is sensitive to small dimensions of [5] M. Belkin, P. Niyogi, V. Sindhwani, Manifold regularization: a
geometric framework for learning from labeled and unlabeled examples,
space at the projection of the matrix. Journal of Machine Learning Research, 7, 2399–2434, 2006
In the future, we will analyze the influence of data [6] D. Donoho, Compressed sensing, IEEE Trans. Info. Theory, vol. 52, no.
normalization on classification rates for dimensionality 4, pp. 1289–1306, September 2006
reduction methods. [7] E. J. Candes, J. Romberg, Quantitative robust uncertainty principles and
optimally sparse decompositions, Foundations of Computational
REFERENCES Mathematics, vol. 6, no. 2, pp. 227–254, April 2006.
[1] L.J.P. van der Maaten, E. O. Postma , H. J. van den Herik, [8] https://www.physionet.org/content/mitdb/1.0.0/ (18 september 2019)
Dimensionality Reduction: A Comparative Review, 2008 [9] M. Fira, L. Goras, N. Cleju, C. Barabasa, On the classification of
[2] Xiaofei He, Partha Niyogi, Locality Preserving Projections, Advances in compressed sensed signals, ISSCS 2011, Iasi, 2011
Neural Information Processing Systems, volume 16, page 37, [10] M. Fira, L. Goras, Dimensionality Reduction for ECG Signals;
Cambridge, MA, USA, The MIT Press, 2004 Laplacian Eigenmaps and Locality Preserving Projections, ISSCS 2019,
[3] Xiaofei He, Locality Preserving Projections. Ph.D. Dissertation. Iasi, 2019
University of Chicago, Chicago, IL, USA. Advisor(s) Partha Niyogi. [11] M. Fira, L. Goras, M. Fira, L. Goras, On Biomedical Signals
AAI3195015. Dimensionality Reduction with Laplacian Eigenmaps, ISMAC - CVB
India, 2019
331 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The volume of internet users is increasingly Fraud detection that has developed very rapidly is fraud
causing transactions on e-commerce to increase as well. We detection on credit cards ranging from fraud detection using
observe the quantity of fraud on online transactions is increasing machine learning to fraud detection using deep learning [6]
too. Fraud prevention in e-commerce shall be developed using but unfortunately fraud detection for transactions on e-
machine learning, this work to analyze the suitable machine commerce is still small, fraud detection research on e-
learning algorithm, the algorithm to be used is the Decision Tree, commerce commerce is still not much so far, fraud detection
Naïve Bayes, Random Forest, and Neural Network. Data to be research on e-commerce is only limited to the determination of
used is still unbalance. Synthetic Minority Over-sampling features or attributes [7] which will be used to determine the
Technique (SMOTE) process is to be used to create balance data.
nature of fraud or non-fraud transactions in e-commerce.
Result of evaluation using confusion matrix achieve the highest
accuracy of the neural network by 96 percent, random forest is The dataset used in this paper has a total of 151,112
95 percent, Naïve Bayes is 95 percent, and Decision tree is 91 records, the dataset classified as fraud is 14,151 records, the
percent. Synthetic Minority Over-sampling Technique (SMOTE) ratio of fraud data is 0.093 percent. Datasets that have very
is able to increase the average of F1-Score from 67.9 percent to small ratios result in an imbalance of data. Imbalance data
94.5 percent and the average of G-Mean from 73.5 percent to results in accuracy results that are more inclined to majority
84.6 percent. data than minority data. The dataset used results more in the
classification of the majority of non-fraud than fraud. Accuracy
Keywords—Machine learning; random forest; Naïve Bayes;
SMOTE; neural network; e-commerce; confusion matrix; G-Mean;
results that are more inclined to majority data make the
F1-score; transaction; fraud classification results worse; handling imbalance data using the
SMOTE (Synthetic Minority Oversampling Technique).
I. INTRODUCTION Recent research about fraud detection in e-commerce
Insight of previous research results on internet users in transactions still determine feature extraction [8], purpose of
Indonesia as released on October 2019 edition of Marketeers this paper is to find the best model to detect fraud in e-
Magazine [1], according to the research the number of internet commerce transactions.
users in Indonesia on 2019 alone, had reached 132 million
users, an increase from the previous year at 143.2 million users In this paper research fraud transaction in ecommerce,
show in Fig. 1. research use dataset from Kaggle, improve classification
machine learning using SMOTE, SMOTE using to handling
The increasing number of internet users in Indonesia has unbalance data, after using SMOTE, dataset will be training
triggered market players in Indonesia to try opportunities to using machine learning. Machine learning is decision tree,
develop their business through internet media. One method Naïve Bayes, random forest, and neural network machine
used is to develop an E-Commerce business [3]. learning to determine accuracy, precision, recall, G-mean, F1-
Score.
Based on statistical data obtained by Statista.com, it is
shown that the number of retail e-Commerce (electronic
commerce) sales in Indonesia will grow 133.5% to the US $
16.5 billion or around IDR 219 trillion in 2022 from the
position in 2017. This growth is supported by the rapid
advances in technology that provide convenience for
consumers to shop.
Huge number of transactions in e-commerce raises the
potential for new problems namely fraud in e-commerce
transactions shows in Fig. 2. The number of e-commerce-
related frauds has also increased every year since 1993. As per
a 2013 report, 5.65 cents lost due to a fraud of every $ 100 in e-
commerce trading turnover. Fraud has reached more than 70
trillion dollars until 2019 [5]. Fraud detection is one way to
reduce the amount of fraud that occurs in e-commerce Fig. 1. Growth of Internet users [2].
transactions.
332 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
333 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
(1)
Where:
B: Data with unknown classes
334 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
E. Neural Network
The algorithm neural network is an artificial intelligence
method whose concept is to apply a neural network system in
the human body where nodes are connected to each other, Fig. 6. Architecture of Neural Network.
architecture neural network as shown in Fig. 6.
F. Confusion Matrix
The number of input layers before training is 11 input
layers, after preprocessing the input layer to 17 input layers, in Confusion matrix is a method that can be used to evaluate
addition to determining the hidden layer, genetic algorithms on classification performance. Table I shows a dataset with only
the neural network is used [24]. The GA-NN [25] algorithm two types of classes [26].
process for this forecasting process is as follows: True Positive (TP) and True Negative (TN) are the number
This forecasting is as follows: of positive and negative classes that are classified correctly,
False Positive (FP) and False Negative (FN) is the number of
Initialization count = 0, fitness = 0, number of cycles positive and negative classes that are not classified correctly.
Based on the confusion matrix, performance criteria such as
Early population generation. Individual chromosomes Accuracy, Precision, Recall, F-Measure, G-Mean can be
are formulated as successive gene sequences, each determined.
encoding the input.
Accuracy is the most common criteria for measuring
Suitable network design classification performance, but if working in an imbalanced
Assign weights class, this criterion is not appropriate because the minority
class will have a small contribution to the accuracy criteria.
Conduct training with backpropagation Looks for The recommended evaluation criteria are recall, precision F-1
cumulative errors and fitness values. Then evaluated Score and G-Mean. F-1 Score is used to measure the
based on the value of fitness. classification of minority classes in unbalanced classes, and the
G-mean index is used to measure overall performance (overall
If the previous fitness <current fitness value, save the
classification performance).
current value
In this study, classification performance using Recall,
Count = count +1
Precision, F-1 Score and G-Mean:
Selection: Two mains are selected using a wheel
roulette mechanism (2)
335 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. CONFUSION MATRIX Decision tree without SMOTE produce Accuracy is 91%,
recall is 59.8%, Precision is 54.1%, F1-Score is 56.8%, G-
Class Predictive Positive Predictive Negative
Mean is 75.2%. Table II shows result from confusion matrix
Actual Positive TP TN decision tree without SMOTE.
Actual Negative FP FN Decision tree with SMOTE produce Accuracy is 91%,
recallis 60.4%, Precisionis 91.6%, F1-Score is 91.2%, G-Mean
IV. RESULTS AND DISCUSSION is 75.3%. Table III shows result from confusion matrix
decision tree with SMOTE.
A. Dataset
This study uses an e-commerce fraud dataset sourced from C. Naïve Bayes
Kaggle. The dataset consists of 151,112 records, a dataset The process of testing using the Naïve Bayes model is done
classified as fraud is 14,151 records, and the ratio of fraud data by preparing data that has already been done in the
is 0.093. SMOTE (Synthetic Minority Oversampling preprocessing process. After preprocessing, the data will be
Technique) [27] minimizes class imbalance in the fraud carried out oversampling using Naïve Bayes classification will
transaction dataset by generating synthesis data, so that the be done using data that has been oversampling, and also Naïve
total data consists of 151,112 records, dataset classified as Bayes will be done using data that is not oversampling. The
fraud is 14,151 records, fraud data ratio is 0.093, as shown in results of these two experiments will show the results of the
Fig. 7. classification using the comparison of Naïve Bayes and the
SMOTE (Synthetic Minority Oversampling Technique)
After oversampling at the picture Fig. 8 oversampling process.
The SMOTE (Synthetic Minority Oversampling Naïve Bayes without SMOTE produce Accuracy is 95%,
Technique) process makes the synthesis data so that the data recall is 54.1%, Precision is 91.1%, F1-Score is 67.9%, G-
becomes balance. Mean is 73.3%. Table IV shows result from confusion matrix
B. Decision Trees naïve Bayes without SMOTE.
The experimental process using the decision tree model is Naïve Bayes with SMOTE produce Accuracy is 95%,
done by preparing data that has been done by the preprocessing recall is 54.2%, Precision is 94.9%, F1-Score is 94.5%, G-
process. After preprocessing, the data will be carried out by Mean is 73.4%. Table V shows result from confusion matrix
oversampling the classification using the decision tree will be Naïve Bayes with SMOTE.
done using the oversampling data, and also the decision tree
will be done by using the data that has not been oversampled. TABLE. II. CONFUSION MATRIX DECISION TREE WITHOUT SMOTE
The results of these two experiments will show the results of
the classification using a comparison of decision trees and the Class Predictive Positive Predictive Negative
SMOTE (Synthetic Minority Oversampling Technique)
Actual Positive 38782 38782
oversampling process.
Actual Negative 1746 2595
336 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
D. Random Forest TABLE. IX. CONFUSION MATRIX NEURAL NETWORK WITH SMOTE
The trial process using the Random Forest model is carried Class Predictive Positive Predictive Negative
out by preparing data that has already been done by the
preprocessing process. After preprocessing, the data will be Actual Positive 38566 2539
carried out classification oversampling using Random Forest Actual Negative 9585 31487
will be done using data that has been oversampled, and also
Random Forest will be done using data that is not Experiments using several algorithms produce accuracy
oversampling. The results of these two experiments will show values as shown in Fig. 9. The highest accuracy value in the
the classification results using the Random Forest comparison neural network algorithm is 96%.
and the SMOTE (Synthetic Minority Oversampling Technique)
oversampling process. Experiments using several algorithms produce recall values
as shown in Fig. 10, recall values increase using machine
Random forest without SMOTE produce Accuracy is 95%, learning algorithms and also the Synthetic Minority Over
recall is 55%, Precision is 95.5%, F1-Score is 69.8%, G-Mean Sampling Technique (SMOTE) compared only using the
is 74.0%. Table VI shows result from confusion matrix random decision tree algorithm, random forest, Naïve Bayes, and
forest without SMOTE. neural networks only, the highest increase occurred in the
neural network algorithm and the SMOTE (Synthetic Minority
Random Forest with SMOTE produce Accuracy is 95%,
Over Sampling Technique).
recall is 58.1%, Precision is 80.5%, F1-Score is 94.3%, G-
Mean is 75.7%. Table VII shows result from confusion matrix Experiments using several algorithms produce precision
random forest with SMOTE. values as shown in Fig. 11, the value decreases using machine
learning algorithm and the Synthetic Minority Over Sampling
E. Neural Network Technique (SMOTE) compared only using the decision tree
Research using the Neural Network model is done by algorithm, random forest, Naïve Bayes, and neural networks,
preparing data that has already been done by the preprocessing highest occurs in neural network algorithms and SMOTE
process. After preprocessing, the data will be carried out (Synthetic Minority Over Sampling Technique).
classification oversampling using Neural Network will be done
using data that has been oversampling, and also Random Forest Experiments using several algorithms produce F1-Score
will be done using data that is not oversampling. The results of values as shown in Fig. 12, F1-Score values are increased by
these two experiments will show the results of classification using machine learning algorithms and also Synthetic Minority
using the Neural Network comparison and the SMOTE Over Sampling Technique (SMOTE) compared only using
(Synthetic Minority Oversampling Technique) oversampling algorithms. F1-Score is used to measure the classification of
process. minority classes in unbalanced classes.
337 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
338 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[10] Srivastava, Abhinav, et al. "Credit card fraud detection using hidden [19] Sadaghiyanfam, Safa, and Mehmet Kuntalp. "Comparing the
Markov model." IEEE Transactions on dependable and secure Performances of PCA (Principle Component Analysis) and LDA (Linear
computing 5.1 (2008): 37-48. Discriminant Analysis) Transformations on PAF (Paroxysmal Atrial
[11] Lakshmi, S. V. S. S., and S. D. Kavilla. "Machine Learning For Credit Fibrillation) Patient Detection." Proceedings of the 2018 3rd
Card Fraud Detection System." International Journal of Applied International Conference on Biomedical Imaging, Signal Processing.
Engineering Research 13.24 (2018): 16819-16824. ACM, 2018.
[12] Aljarah, Ibrahim, Hossam Faris, and Seyedali Mirjalili. "Optimizing [20] Harrison, Paula A., et al. "Selecting methods for ecosystem service
connection weights in neural networks using the whale optimization assessment: A decision tree approach." Ecosystem services 29 (2018):
algorithm." Soft Computing 22.1 (2018): 1-15. 481-498.
[13] Bouktif, Salah, et al. "Optimal deep learning lstm model for electric load [21] Randhawa, Kuldeep, et al. "Credit card fraud detection using AdaBoost
forecasting using feature selection and genetic algorithm: Comparison and majority voting." IEEE access 6 (2018): 14277-14284.
with machine learning approaches." Energies 11.7 (2018): 1636. [22] Lakshmi, S. V. S. S., and S. D. Kavilla. "Machine Learning For Credit
[14] Xuan, Shiyang, Guanjun Liu, and Zhenchuan Li. "Refined weighted Card Fraud Detection System." International Journal of Applied
random forest and its application to credit card fraud detection." Engineering Research 13.24 (2018): 16819-16824.
International Conference on Computational Social Networks. Springer, [23] Li, Tong, et al. "Differentially private Naïve Bayes learning over
Cham, 2018. multiple data sources." Information Sciences 444 (2018): 89-104.
[15] Hong, Haoyuan, et al. "Landslide susceptibility mapping using J48 [24] Suganuma, Masanori, Shinichi Shirakawa, and Tomoharu Nagao. "A
Decision Tree with AdaBoost, Bagging and Rotation Forest ensembles genetic programming approach to designing convolutional neural
in the Guangchang area (China)." Catena 163 (2018): 399-413. network architectures." Proceedings of the Genetic and Evolutionary
[16] Zhao, Jie, et al. "Extracting and reasoning about implicit behavioral Computation Conference. ACM, 2017.
evidences for detecting fraudulent online transactions in e-Commerce." [25] Ruehle, Fabian. "Evolving neural networks with genetic algorithms to
Decision support systems 86 (2016): 109-121. study the string landscape." Journal of High Energy Physics 2017.8
[17] Sharma, Shiven, et al. "Synthetic oversampling with the majority class: (2017): 38.
A new perspective on handling extreme imbalance." 2018 IEEE [26] Ting, Kai Ming. "Confusion matrix." Encyclopedia of Machine Learning
International Conference on Data Mining (ICDM). IEEE, 2018. and Data Mining (2017): 260-260.
[18] Kim, Jaekwon, Youngshin Han, and Jongsik Lee. "Data imbalance [27] Siringoringo, Rimbun. "Klasifikasi Data Tidak Seimbang Menggunakan
problem solving for smote based oversampling: Study on fault detection Algoritma Smote Dan K-Nearest Neighbor." Journal Information
prediction model in semiconductor manufacturing process." Advanced System Development (ISD) 3.1 (2018).
Science and Technology Letters 133 (2016): 79-84.
339 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Every year business is overwhelmed by the these difficulties and complexities to interpret this data [4] [5]
quantity and variety of data. Visualization of Multi-dimensional [6] [7]. This interference could lead to a complexity in reading
data is counter-intuitive using conventional graphs. Parallel or interpreting the data.
coordinates are proposed as an alternative to explore
multivariate data more effectively. However, it is difficult to Previous research has proposed exploratory techniques to
extract relevant information through the parallel coordinates enhance the visualization of multidimensional data. Within the
when the data are Multi-dimensional with thousands of lines last 20 years researches focused on Techniques to reduce the
overlapping. The order of the axes determines the perception of number of poly-lines or reducing or reordering the parallel axes
information on parallel coordinates. This paper proposes three [8] [9]. This paper introduces novel techniques for reordering
new techniques in order to arrange the axes in the most the factors of the data based on the correlation coefficient
significant relation between the datasets. The datasets used in calculations. The goal of these techniques is to facilitate the
this paper, for Egyptian patients, with many external factors and readiness and the complexity of the parallel coordinates. The
medical tests. These factors were collected by a questionnaire paper categorized into different sections, the proposed
sheet, made by medical researchers. The first Technique methods, a detail explanation about the new techniques
calculates the correlation between all features and the age of the proposed. The results and discussion the comparison between
patient when they get diabetes disease. The second technique is the three techniques and finally the conclusion section.
based on merging different features together and arranging the
coordinates based on the correlations values. The Third II. BACKGROUND AND RELATED WORK
Technique calculates the entropy value for each feature and then
arrange the parallel coordinates in descending order based on The Parallel coordinates is an interactive visualization, and
the positive or negative values. Finally based on the result is the most used for multidimensional data visualization. It was
graphs, we conclude that the second method was more readable developed and popularized by Alfred Inselberg [10].
and valuable than the other two methods. Improving the parallel coordinates plot is a highly active
research topic. There are some techniques proposed in previous
Keywords—Parallel coordinates; visualization; correlation research that attempted to enhance the readability of the results
coefficient; entropy function by applying clustering techniques or sampling polynies [11]
[12] [13] [14]. Moreover the readiness and effectiveness of the
I. INTRODUCTION parallel coordinates depends on ordering the dimensions and
In the recent studies of computer science and technologies, factors, different dimension ordering techniques were
an accelerating information explosion is being witnessed. In presented [15] [16] [17].
digital universe today about 2.7 Zeta bytes executed
Other papers proposed new methods for interpreting the
continuously [1]. Based on the Estimations and studies
readiness of the parallel coordinate by dividing the dimensions
presented by the International Data Corporation (IDC), they
of the datasets input into groups of lower dimensions based on
suspect that by 2020 business transactions on the internet-
the correlations calculations; the conclusion of this technique
business-to-business and business-to-consumer will reach 450
can represent various groups of correlated dimensions in high
billion per day [2]. Moreover, analysis and knowledge are
dimensional data space [8].
power and in order to analysis and interpret these huge
amounts of data, Users have to use tools to visualize this data. Furthermore, another paper proposed the automated
These visualization tools can assist in retrieving valuable assistance to rearrange the order of the variable; this
information, which may effectively help in solving many automation was done using a system called V-miner. Motorola
different types of problems. One of these important tools is the engineers were affected by the new powerful enhancements
Parallel coordinates, method of visualizing high dimensional and also facilitate the use of the parallel coordinates [4].
geometry and analyzing multidimensional data [3].
Also techniques were proposed to simplify the
These days the data and its dimensions' increase rapidly representation of the parallel coordinate visualization, where a
which results too much interference in the coordinates and new study proposed using the eye tracking. The main idea is to
timelines of the parallel coordinates, lead to obstacles in understand whether the parallel coordinate visualizations are
analyzing. For this reason, many papers are presented to solve easy to be perceived or not.
340 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
From the results of this study, the users were able to A. Datasets
interpret and realize the parallel coordinate easily by The Egyptian National Research Center compiled these
concentration on the correct areas for the chart [18]. Datasets based on a medical questionnaire which contains 348
III. THE PROPOSED METHODS patients. This questionnaire is comprised of questions
regarding the risk factors that cause diabetes disease and were
This section will discuss the proposed methods to enhance questions for diabetes patients. After that these forms were
the visualization in the parallel coordinates. The goal of using extracted into a statistical tool called SPSS, for doing statistical
the Parallel coordinates is one of the most important techniques analysis on this data and finally they were exported into an
to visualize dataset with multidimensional datasets, the better excel sheet, in order to be used in experiments as shown in
visualization becomes obvious, and more information can be Fig. 3.
retrieved [19]. The results of the parallel coordinate
visualization always confuse the reader, and could lead to The datasets were collected 6-years ago. The Dataset
difficulties to read. Past studies proved that the correlation comprised of 23 features; these features are summarized in
coefficient affects the result and the interpretation of the Table I.
parallel coordinate visualization [20]. The effectiveness on the
TABLE. I. DESCRIPTION OF DATASET FEATURE
interpretation and the readiness, also has an effect on the
visualization between two coordinates, for instance, the parallel No. Feature name Type Range
coordinates plot for data that have negative 1 correlation 1 Diabetes Age Numeric Real Values
different from the parallel coordinates for data that have 1
2 Gender Numeric Categorical
correlation is as follows:
3 Education Numeric Categorical
In Fig. 1 and 2, the correlation affects the visualization of 4 Diabetic Family member Numeric Categorical
the parallel coordinate chart, and incase the two features are 5 Smoker Numeric Categorical
correlated or not the lines interfere or move in parallel path.
6 Cigarette number Numeric Real Values
For this reason, this paper proposed two of the new methods
based on the correlation coefficient. In order to simplify the 7 Smoking Start Date Date Real Values
complexity of the intervention between lines that may lead to 8 Exercising Status Numeric Categorical
difficulties in tracking the parallel coordinate’s graphs. 9
Frequent Exercise per
Numeric Real Values
week
Moreover, these techniques give the user a better chance to 10 Exercise Type Numeric Categorical
interpret and analyze the datasets more professionally. The 11 Food Type Numeric Categorical
used datasets are for Egyptian people suffering from the
12 Healthy Food status Numeric Categorical
diabetes disease. This data was collected by the Egyptian
National Research Center and was based on standard medical 13 No of Basic Meals Numeric Real Values
questionnaire. This questionnaire was prepared by specialized 14 Snacks Status Numeric Categorical
doctors in the diabetic field. 15 Snacks Number Numeric Real Values
The goal of implementing these two methods on the 16 Snack Type Numeric Categorical
diabetic patients’ dataset is to reach the most significant 17 Regime Status Numeric Categorical
features that affect the health of these patients and assist in 18 Blood Pressure Status Numeric Categorical
triggering the diabetic disease faster in younger ages. 19 Blood Fat Status Numeric Categorical
20 Foot Complications Numeric Categorical
21 Neuro Complications Numeric Categorical
22 Low Vision status Numeric Categorical
23 Wound Recovery Status Numeric Categorical
Fig. 2. Parallel Coordinates Plot for Data with Negative Correlation Fig. 3. The Correlation Coefficient for Each Variable with Respect to the
Coefficient. Age.
341 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
342 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 8. Parallel Coordinates Chart based on the 1st Method Calculating the
Correlation for Each Feature with Respect to Age Feature.
Fig. 11. 1st Method Focus on the Education Feature and Specially the Highest
Level of Education.
Fig. 9. Parallel Coordinates Chart based on the 2nd Method Calculating the
Correlation by Merging the Previous Features with the Respect to Age Fig. 12. 2nd Method Focus on the Education Feature and Specially the Highest
Feature. Level of Education.
Fig. 10. Parallel Coordinates Chart based on the 3rd Method Calculating the Fig. 13. 3rd Method Focus on the Education Feature and Specially the
Entropy Function for Each Feature then Rearrange the Coordinates Highest Level of Education.
Accordingly.
343 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
REFERENCES
[1] G. Noseworthy, "Infographic: Managing the Big Flood of Big Data in
Digital Marketing," [Online]. Available: http://analyzingmedia.com/2012/
infographic-big-flood-of-big-data-in-digitalmarketing/.
[2] V. T. M. S. Carrie MacGillivary, "IDC's Worldwide Internet of Things
Taxonomy," IDC, 2015.
[3] M. G. B. Akbar, "Data Analytics Enhanced Data Visualization and
Interrogation with Parallel Coordinates Plots," in 26th International
Conference on Systems Engineering, ICSEng 2018 , 2019.
[4] T. M. T. A. S. Kaidi Zhao, "Detecting Patterns of Change Using
Enhanced Parallel Coordinates Visualization," in ICDM '03 Proceedings
of the Third IEEE International Conference on Data Mining, ,2003.
[5] X. Y. Z. G. X. Huamin Qu, "Scattering Points in Parallel Coordinates,"
IEEE Transactions on Visualization & Computer Graphics, vol. 15, pp.
Fig. 14. 1st Method Focus on Smoking Feature and Specially the Smokers. 1001-1008,, 2009.
[6] W. Sun and S. Wang, "A new data mining method for early warning
landslides based on parallel coordinate," in Proceedings 2011 IEEE
International Conference on Spatial Data Mining and Geographical
Knowledge Services, 2011.
[7] G. R. ,. T. J. ,. F. L. D. A. a. R. B. Joris Sansen *, "Visual Exploration of
Large Multidimensional Data Using Parallel Coordinates on Big Data
Infrastructure," Informatics, vol. 7, 2017.
[8] Z. Y. T. I. F. Haruka Suematsu, "Arrangement of Low-Dimensional
Parallel Coordinate Plots for High-Dimensional Data Visualization," in
2013 17th International Conference on Information Visualisation, 2013.
[9] K. Zhao, B. Liu, T. Tirpak and A. Schaller, "Detecting Patterns of
Change Using Enhanced Parallel Coordinates Visualization," in Third
IEEE International Conference on Data Mining, 2003.
[10] B. D. Alfred Inselberg, "Parallel coordinates: a tool for visualizing multi-
Fig. 15. 2nd Method Focus on Smoking Feature and Specially the Smokers. dimensional geometry," in Proceedings of the First IEEE Conference on
Visualization: Visualization `90, 1990.
[11] M. W. E. R. Y. Fua, "Hierarchical parallel coordinates for exploration of
large datasets," in Proceedings Visualization '99 (Cat. No.99CB37067),
1999.
[12] P. L. M. J. M. C. J. Johansson, "Revealing Structure within Clustered
Parallel Coordinates Displays,," in INFOVIS '05 Proceedings of the
Proceedings of the 2005 IEEE Symposium on Information Visualization,
2005.
[13] F. C. P. A. A. M. V. Elena Geanina ULARU, "Perspectives on Big Data
and Big Data Analytics," Database Systems Journal, Vols. vol. III, no.
4/2012, pp. 3-14, 2012.
[14] X. Y. H. Q. W. C. B. C. H. Zhou, "Visual Clustering in parallel
Coordinates," in EuroVis'08 Proceedings of the 10th Joint Eurographics /
IEEE - VGTC conference on Visualization, 2008.
Fig. 16. 3rd Method Focus on Smoking Feature and Specially the Smokers. [15] S. T. S. J. Hemant Mekwana, "Axes Re-ordering in parallel coordinate for
pattern Optimization," International Journal of Computer Applications ,
vol. Volume 40– No.13, pp. 42-47, 2012.
V. CONCLUSION [16] L. F. Lu, M. L. Huang and T.-H. Huang, "A New Axes Re-ordering
In this paper, three techniques to reorder the coordinates of Method in Parallel Coordinates visualization," in 11th International
the charts were introduced. Two of these techniques based on Conference on Machine Learning and Applications, 2012.
the correlation coefficient and the third one based on the [17] J. Z. B. H. R. Rosenbaum, "Progressive Parallel Coordinates," in IEEE
Pacific Visualization Symposium, 2012.
entropy function. The goals of these techniques to enhance the
[18] H. Siirtola, T. Laivo, T. Heimonen and K.-J. Räihä, "Visual Perception of
parallel coordinate visualization and facilitate the interpretation Parallel Coordinate Visualizations," in 13th International Conference
of data. Information Visualisation, 2009.
Concluding based on the analysis and by comparison, the [19] Tran Van Long, "Visualizing High-density Clusters in Multidimensional
Data," Jacobs University, 2009.
second method results a better visualization than others. New
information was interpreted and extract from the charts. In the [20] J.-B. M. a. J. J. V. W. J. Li, "Judging correlation from scatter plots and
parallel coordinate plots," Information Visualization, vol. Volume 9 Issue
future work a plan to merge between the three techniques with 1, pp. 13-30, 2010.
the clustering methodology. Moreover, further analysis and [21] C. Kamath, Scientific Data Mining : A practical Perspective,, Society for
discussion will be held between the old and the new charts. Industrial and Applied Mathematics, 2009.
344 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Location is one of the most important factors to for job search but their tools are specifically designed for
consider when looking for offering a new job. Currently, there American job seekers only and they need to be personalized.
exist many job recommender systems to help match the right They require recommender systems to meet job seekers
candidate with the right job. A review of the existing preferences as thousands of jobs are posted on these websites
recommender systems, included within this article, reveals that daily, and it takes a great deal of effort to find the right
there is an absence of appropriate mapping support offering for position.
job recommendation. This article aims to propose a general map-
based job recommender model, which is implemented and Currently, there exist many personalized job search
applied within a system for job seekers in Saudi Arabia. The systems, i.e. job recommender systems to help match the right
system adapts content-based technique to recommend jobs using candidate with the right job. Examples include CASPER [1],
the cosine similarity and will help Saudi job seekers finding their Proactive [2], FES [3], PROSPECT [4], eRecruiter [5], iHR
desired job in an efficient way using interactive maps. This [6], RésuMatcher [7] and the work of [8]. The work of [9] and
ultimately will contribute to Saudi Arabia moving forward to the [10] provides comprehensive review on job recommender
digital transformation which is one of the major objectives to systems.
fulfill the Saudi vision 2030.
Table I shows a general comparison among these systems
Keywords—Recommender systems; content-based based on the following characteristics: recommendation input,
recommendation; location-based search; maps recommendation technique, and the offering of mapping
support in these systems.
I. INTRODUCTION
All the recommender systems of Table I support
Finding a job in today‟s market is a major challenge. A
personalization in the job search, although they differ in the
common way to look for a job is to use job search websites.
recommendation technique used. Most of the system use
Rather than taking the time to search newspapers, company
collaborative filtering recommendation (CFR) and content-
web sites, and other traditional job postings. A job search
based recommendation (CBR). Unfortunately, most of these
website can do it all with the click of a button. A job search
systems are poor in personalized search because their search
engine facilitates the matching and communication of job
functionality is limited to keyword-based search, often
opportunities between job seekers and employers.
resulting in poor, irrelevant search results. For example, a job
The location of a job has the potential to significantly affect search using the keyword “Java” to search for jobs within a
an individual‟s lifestyle. People often tend to focus their job limited geographical location (New York, NY) on
search on a particular area and the job location can play an www.indeed.com returned over 8000 jobs.
important part in the decision to apply for a job. Thus, location
In the context of this research, it is worth mentioning that
is one of the most important factors to consider when looking
none of them offer mapping support tools. There are a number
for a new job. As such, every job search website worth using
of international job search web sites, including Indeed
has the ability to search for jobs based on location.
(www.indeed.com), Monster (www.monster.com), Glassdoor
In the most popular job search websites, e.g. Indeed (www.glassdoor.com),CareerBuilder(www.careerbuilder.com),
(www.indeed.com) and Monster (www.monster.com), the SimplyHired(www.simplyhired.com),Pathwayjobs(www.path
representation for the retrieved information may not be wayjobs.com), and LinkedIn (www.linkedin.com), that provide
appropriate for job seekers with respect to the job location. The geospatial search. To the best of our knowledge, only two of
reason is because the results returned to a job seeker about the these websites, namely Glassdoor and Pathwayjobs, provide
job locations are in textual form with no provision of a map for mapping tools for job search and both of them are for
the employers‟ geospatial location. American job seekers only. Anyway, none of them are
recommender systems.
As such, users of such websites may find difficult searching
for their desired job, and not efficient because the lack of the This research aims is to bridge the above two mentioned
website support in displaying the available jobs on a map gaps in the current job search websites and systems, and
might lead to them spending a significant amount of their time proposes a personalized model based on job seeker preferences
reading and reviewing their options regarding job location. and also support location-based search with interactive
mapping tool. The proposed model and prototype will help job
Two job search websites, namely Glassdoor seekers, especially in Saudi Arabia, to find the right job that
(https://www.glassdoor.com/Job/explorer/index.htm) and meets their qualifications in an efficient way using interactive
Pathwayjobs (www.pathwayjobs.com), provide mapping tools maps.
345 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Personal information
User behaviour (revisited data, read time data,
CFR
CASPER 2000 [1] activity data, feedback) No
CBR
Query (job description, salary, location, education
background)
CBR
PROSPECT 2010 [4] Personal information No
(Resume Mining)
Personal information ( include interest) CBR
eRecruiter 2011 [5] No
User behaviour ( include feedback in real-time) KBR (Ontology)
CBR
Personal information CFR
iHR 2013 [6] No
User behaviour Hybrid
Cluster-based
Social JRS 2013 [11] Users data, users friends data, users profile SVM No
Lucene IKAnalyzer
iHR+ 2015 [12] Basic information only No
TF-ADF
SKILL 2015 [13] Skills in a resume , requirements in job poster Skills taxonomy, skills tagging No
Statistical similarity
RésuMatcher 2016 [7] Resume document No
FST
346 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
347 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
348 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
C. Location Geocoder The results got form; this module is shown in Fig. 7. To
Geocoding is the process of converting addresses into implement this module, Google Geocoding API is used. In
geographic coordinates, i.e. latitude and longitude. order to use Google APIs, an API key is needed. This key
should be later embedded in the code.
This module reads the addresses of the recommended jobs
and formats the address to be complete if they are not. The D. Geometry Visualizer
geographic coordinates of the formatted addresses are then Markers are one of the most common ways to visualize
found. As an example, the address of the third recommended locations on a map. In this module, the geocoded addresses of
job in Fig. 6 was extracted as: the recommended jobs are visualized on a map using markers.
Fig. 8 shows on a map the locations of the six recommended
Al Othman Agriculture Production and Processing jobs shown in Fig. 6.
Company (NADA), Khobar, Saudi Arabia
To implement this module, Google Maps JavaScript API is
In order to be able to visualize the address on a map, the used.
Location Geocoder module finds the complete address, i.e.
formatted address as well as the geographic coordinates, i.e.
geometry, in terms of latitude and longitude.
349 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 7. Geocoded Address of a Selected Recommended Job, Arrows Shows the Formatted Address, Latitude and Longitude of a given Address.
IV. A COMPARISON BETWEEN THE MAP-BASED JOB on job seeker‟s qualifications and preferences. However, in our
RECOMMENDER PROTOTYPE AND OTHER SAUDI JOB SYSTEMS prototype, the location is highly considered during the
recommendation process. We believe that, in Saudi Arabia, job
As an initial evaluation for the proposed model, the location is essential when looking for a new job. This is due to
prototype that has been implemented based on that model is the large geographical area of the country and the long
compared with JADARAH. The most commonly used system distances among its several regions. Unlike our system,
by Saudi job seekers. JADARAH was developed on 2011 by JADARAH does not pay great attention to job seeker location.
the ministry of civil service to help Saudi job seekers find jobs That is, a person who lives in Qassim which is located in the
within Saudi government sectors. The comparison is made in central region may be got a job in Abha which is located in the
terms personalization, map-based support, and type of jobs southern region.
offered.
The proposed system is map-based as the recommended
For personalization criterion, both systems are jobs are displayed on a map. Unlike our system, JADARAH
recommender systems. That is, the jobs are personalized based displays the recommended job in textual form.
350 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In terms of types of jobs, our system considers jobs from Recommender,” in Third International Conference on Autonomic and
private companies as well as governmental sectors. On the Autonomous Systems (ICAS‟07), 2007, pp. 21–21.
other hand, JADARAH is only for governmental jobs. [3] J. O. Daramola, O. O. Oladipupo, and A. G. Musa, “A fuzzy expert
system (FES) tool for online personnel recruitments,” Int. J. Bus. Inf.
It is worth mentioning that there are other job requirement Syst., vol. 6, no. 4, p. 444, 2010.
systems in Saudi Arabia such as bayt.com, wadhefa.com, and [4] A. Singh, C. Rose, K. Visweswariah, V. Chenthamarakshan, and N.
tanqeeb.com, however, these systems were not included in the Kambhatla, “PROSPECT: A system for screening candidates for
recruitment,” in Proceedings of the 19th ACM international conference
comparison as these systems do not provide personalized on Information and knowledge management - CIKM ‟10, 2010, p. 659.
search, i.e. not recommender systems. [5] M. Hutterer, “Enhancing a Job Recommender with Implicit User
Feedback,” vol. 2011, 2011.
V. CONCLUSION
[6] W. Hong, S. Zheng, H. Wang, and J. Shi, “A Job Recommender System
Although there are many job recommendation systems, Based on User Clustering,” J. Comput., vol. 8, no. 8, Aug. 2013.
these systems do not offer mapping support. In line with the [7] S. Guo, F. Alamudun, and T. Hammond, “RésuMatcher: A personalized
digital transformation objectives of the new Saudi vision 2030, résumé-job matching system,” Expert Syst. Appl., vol. 60, pp. 169–182,
and in order to improve job search in general and in Saudi Oct. 2016.
Arabia in particular, this work proposed a personalized and [8] W. Shalaby et al., “Help Me Find a Job: A Graph-based Approach for
Job Recommendation at Scale,” Dec. 2017.
map-based job search model. The model was theoretically
[9] Z. Siting, H. Wenxing, Z. Ning, and Y. Fan, “Job recommender systems:
based on the existing recommender systems in the literature A survey,” in 2012 7th International Conference on Computer Science
and used content-based recommendation with integration of & Education (ICCSE), 2012, pp. 920–924.
mapping feature for location-based search which has never [10] S. T. Al-Otaibi and M. Ykhlef, “A survey of job recommender systems,”
been used in the previous systems. The cosine similarity was Int. J. Phys. Sci., vol. 7, no. 29, pp. 5127–5142, Jul. 2012.
used for the content-based recommendation with a minimum [11] M. Diaby, E. Viennet, and T. Launay, “Toward the next generation of
threshold of 0.8 for job title search and further narrowed down recruitment tools: An online social network-based job recommender
by job location comparison. As a proof of concept, a prototype system,” in Proceedings of the 2013 IEEE/ACM International
was implemented based on the proposed model. The proposed Conference on Advances in Social Networks Analysis and Mining,
ASONAM 2013, 2013, pp. 821–828.
system provides better features compared with JADARAH, the
[12] H. Wenxing, C. Yiwei, Q. Jianwei, and H. Yin, “iHR+: A mobile
most common recruitment system used by Saudi job seekers. reciprocal job recommender system,” in 2015 10th International
The proposed system in this research work will help Saudi job Conference on Computer Science & Education (ICCSE), 2015, pp. 492–
seekers finding the desired job in an efficient way using maps. 495.
In this work, the proposed system is meant by Saudi job [13] M. Zhao, F. Javed, F. Jacob, and M. McNair, “SKILL: A system for
seekers as it is based on Mihanti.com as a main source for jobs. skill identification and normalization,” in Proceedings of the National
In the future, we plan make more general and grab jobs from Conference on Artificial Intelligence, 2015, vol. 5, pp. 4012–4017.
several international job‟s boards. [14] S. Maheshwary and H. Misra, “Matching Resumes to Jobs via Deep
Siamese Network,” in WWW ‟18 Companion Proceedings of the The
Web Conference 2018, 2018, pp. 87–88.
ACKNOWLEDGMENT
[15] K. Haruna, M. Akmar Ismail, D. Damiasih, J. Sutopo, and T. Herawan,
This research is funded by the Deanship of Scientific “A collaborative approach for research paper recommender system,”
Research (SRD), Qassim University, Saudi Arabia. The PLoS One, vol. 12, no. 10, p. e0184516, Oct. 2017.
associated project number of the research is [16] Y. Balachander and T.-S. Moh, “Ontology Based Similarity for
coc_2018_1_14_S_5143. Information Technology Skills,” in 2018 IEEE/ACM International
Conference on Advances in Social Networks Analysis and Mining
REFERENCES (ASONAM), 2018, pp. 302–305.
[1] R. Rafter, K. Bradley, and B. Smyth, “Personalised Retrieval for Online [17] Y. Wang, M. Wang, and W. Xu, “A Sentiment-Enhanced Hybrid
Recruitment Services,” 2000. Recommender System for Movie Recommendation: A Big Data
[2] D. H. Lee and P. Brusilovsky, “Fighting Information Overflow with Analytics Framework,” 2018.
Personalized Comprehensive Information Access: A Proactive Job
351 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Several regions of the world have been affected by the other existing approaches with the minimum root mean
one of the natural disasters named as flash floods. Many villagers squared error of 0.0002 and prediction speed of 35000
who live near stream or dam, they suffer a lot in terms of observations per second.
property, cattle and human lives loss. Conventional early
warning systems are not up to the mark for the early warning Keywords—Flash floods; classification; SVM; k-NN; logistic
announcements. Diversified approaches have been carried out regression; quadratic SVN; ensemble bagged trees; exponential
for the identification of flash floods with less false alarm rate. GPR
Forecasting approaches includes some errors and ambiguity due
to the incompetent processing algorithms and measurement I. INTRODUCTION
readings. Process variables like stream flow, water level, water Low cost effective solution has been designed using
color, precipitation velocity, wind speed, wave’s pattern and Android phone and Arduino. Echolocation strategy has been
cloud to ground (CG) flashes have been measured for the robust applied to measure the water level [1]. Bunch of sensors were
identification of flash floods. A vibrant competent algorithm
deployed from the upper stream to the village. Sensors nodes
would be required for the investigation of flash floods with less
were designed to observe the data. Supervisor control data
false alarm rate. In this research paper classifiers have been
applied on the collected data set so that any researcher could
acquisition (SCADA) based system was designed to forecast
easily know that which classifier is competent and can be further the floods on real time basis [2]. Torrential downpour can be
enhanced by combining it with other algorithms. A novel considered as the main cause of flash floods. Heavy
comprehensive parametric comparison has been performed to precipitation for the short time known as torrential downpour.
investigate the classification accuracy for the robust classification Two meteorological radars have been used to observe the
of false alarms. For the better accuracy more than one process torrential downpour in Czech Republic [3]. Ultrasonic sensor
variables have been measured but still contained some false based observations have been analyzed to determine the dam
alarm. Appropriate combination of sensor was integrated to level [4]. Data from 2007 to 2010 were collected from three
increase the accuracy in results as multi-modal sensing device stations of Selangor to design a novel model for the prediction
has been designed to collect the data. Linear discriminant of flash floods using ANN. Feed forward back propagation
analysis, logistic regression, quadratic support vector machine, k- with the tangent sigmoid function was proposed to estimate
nearest neighbor and Ensemble bagged tree have been applied to the floods. The process variables like humidity, rainfall, and
the collected data set for the data classification. Results have temperature were taken as input and the rainfall data from the
been obtained in the MATLAB and discussed in detail in the stations were set as the targets. The proposed model best
research paper. The worst accuracy of the classification (62%) results [4]. Another research elaborated a novel method by
has been achieved by the coarse k-NN classifier that means measuring the magnetic field lines by tesla meter or
coarse k-NN produced 38% false negative rate that is not
magnetometer to predict the flash floods. Research proved by
acceptable in the case of forecasting. Ensemble bagged trees
showing the results that during the flash floods magnetic field
produced best classification results as it achieved 99 % accuracy
and 1% error rate. Furthermore, according to the
line reduced abruptly therefore magnetic field lines that are
comprehensive parametric comparison of regression models radiated from the center of the earth can be regarded as the
Quadratic SVM found to be the worst with mean square error of significant yardstick to measure the flash floods [5]. A
0.5551 and time elapsed 13.159 seconds. On the other hand, practical early warning system must detect the flash floods in
Exponential Gaussian process regression performed better than appropriate timings as there is no use of the system if the flood
352 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
is detected after the starting of the event. Keeping in the mind TABLE. I. COMPARATIVE ANALYSIS OF VARIOUS APPROACHED FOR THE
the time constraint a flood prediction model using hybrid FLASH FLOOD INVESTIGATION [20]
approach NNARX (Neural Network Autoregressive with
Performan ANN-
Exogenous Input) with EKF (extended Kalman Filter) was ce Indices
ANN SVM ANFIS NNARX
PSO
developed. 120 samples of the data set were tested and results
RMSE 0.194 0.390 0.116 0.090 0.0047
showed that hybrid approach worked better [6]. In a previous
method a novel solution was suggested by measuring the Best Fit 73 64 78 80.10 98.7
carbon dioxide levels in the environment and soil flux as the Satisfacto Unsatisfacto Satisfacto Satisfacto Satisfacto
Results
newly discovered phenomena proved that plants take less ry ry ry ry ry
water due to the increased amount of carbon dioxide levels. Hourly
6 hrs 6 hrs 3 hrs 3 hrs 3 hrs
Multi-layer perceptron was applied on the collected data set to Data
reduce the false alarms in determining the flash floods [7]. IT Accuracy 73 64 78 80.16 98.99
based flash flood monitoring was performed for the immediate Precision Medium Low High High High
emergency rescue of the affected people in Jeddah [8].
Majority casualties happen due to the lack of the data and Reliability Medium Low High High High
information regarding the propagation of the flash flood. Power
Limited Limited Limited Limited Limited
Therefore, an urban flash flood monitoring was performed to Utilixation
know the actual and predicted flow of the flash floods for the
evacuation announcement [9]. Kalman filtering, fuzzy logic, II. PROBLEM STATEMENT
clustering, Neural network autoregressive model with It has been highlighted in the literature review that sensors
exogenous input (NNARX), Particle swarm optimization and transducers produce false alarms. Errors are generated
(PSO) and Support vector machine have been applied for the usually in instrumentation and measurement [10]. Sometime
prediction and estimation of flash floods [10]. Fuzzy logic prediction of flash floods can be wrong due to the incompetent
based a disaster management device has been designed for the decision algorithms and poor sensitivity of sensors. Due to the
announcement of exit routes during the hazard [11]. Ensemble increased number of false alarms a competent and vigorous
learning model has also been designed for the better classifier and regression model was required for the
generalization model of classification [12]. The false positive discrimination of true positive vale and false positive value.
rate (FP) is the date values which have been estimated wrong Sensors data values may contain false alarms and missed
due to the presence of error. The negative false rate (FN) is the values [21].
data that is defined falsely as negative [13]. False alarm ratio
relies on the relation of the complexity and anisotropy of the III. MATERIALS AND METHODS
sea-floor Williams et al. [14] [15]. Gaussian process Multi-modal sensing device was developed to collect the
regression model can be acknowledged as the competent data from any sea shore. The data has been collected from the
approach for solving non-linear regression issues. It performs sea shore of Kund Malir, Pakistan. Selection of transducers
regression in a simple way parameterization and Bayesian. It was not an easy task as appropriate combination of competent
also removes ambiguity and uncertainty in the prediction of sensors was needed. According to the literature review almost
event [16]. Data driven approaches are usually capable to all the parameters have been used for the flash floods
determine the complex and non-linear data to be transformed investigation.
for the prediction of the event [17]. Support vector machine
was developed in 1990s and became popular among the A. Multi-Resolution Sensing Device for Data Collection
classification model to its better learning generalization [18]. Fig. 1 shows that a device has been developed for the
Signal attenuation and distortion in Television satellites due to investigation of flash flood as a hazard monitoring device.
the rainfall was observed and flooding was mapped. It has Multi-modal sensing device comprised of the following
been analyzed that Ku band frequencies varied due to rain fall sensors: (a) pressure (b) temperature (c) water level (d) gas
and climate change. Simulated maps of flash floods were sensor for detecting CO2 and (e) ultrasonic sensor. Selection of
compared with the existing mapping methods to validate the sensors was very complex task as bunch of appropriate
approach [22]. sensors must be used for the accurate and precise results
without any false alarm rate [21].
Table I demonstrates the tabular chart for the comparison
of AI based algorithms. Root mean square error has been a B. Fundamental Flow Diagram
yardstick to estimate the performance of the algorithms. The In Fig. 2, Ultrasonic sensor, Passive infrared sensor, MQ2
results of actual run time data using existing approach of sensor, humidity sensor, pressure and temperature sensors
MLP-PSO show probability of 95.15%. The proposed have been used to measure the data near the sea shore of Kund
algorithm of ANN-PSO has performed the investigation of Malir, Pakistan. Data labeling was performed. The data may
flash flood with 0.0047 error probability and enhanced contain random, missed and repetitive values it must be
accuracy [20]. filtered or normalized before the processing. Therefore, robust
classification and regression model were needed.
353 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
354 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
IV. EXTENSIVE PARAMETRIC COMPARISON OF DATA TABLE. IV. PARAMETRIC COMPARISON OF LDA, LR, QSVM, K-NN AND
CLASSIFIERS ENSEMBLE
Table IV demonstrated that Comprehensive parametric True
comparison has been performed to investigate the False Prediction
Classification Positive Accuracy Training
Negative speed per
classification accuracy for the robust classification of false Models rate
rate (%)
(%)
second
Time (s)
alarm in predicting flash floods. Linear discriminant analysis, (%)
logistic regression, quadratic support vector machine, k- Linear
89 11 97.0 12000 1.5238
nearest neighbor and Ensemble bagged tree have been applied Discriminant
to the collected data set for the data classification. Initially Logistic
seventy-five percent of the data were used as a training and 93 7 96.4 15000 3.9633
Regression
other 25 percent data was saved for the testing purpose. Both Quadratic
of the data files training and testing were converted into the 93 7 96.8 40000 0.9327
SVM
variable so that it may utilized in the MATLAB as all the
simulations have been performed in the MATLAB. MATLAB Fine k-NN 96 4 98.6 9600 1.5307
based simulations produced the confusion matrix and all the Medium k-
91 9 96.6 19000 0.927
parametric results which have been presented in the table. The NN
worst accuracy of the classification (62%) has been achieved Coarse k-NN 62 38 89.3 16000 0.90243
by the coarse k-NN classifier that means coarse k-NN
produced 38% false negative rate that is not acceptable in the Ensemble
99 1 99.4 2800 6.4101
case of forecasting. Ensemble bagged trees produced best Bagged Trees
classification results as it achieved 99 % accuracy and 1%
error rate.
A. Linear Discriminant Analysis
Fig. 3 explains that data set was trained for the linear
discriminant classification and it produced confusion matrix
and other results. Confusion matrix showed that LDA
achieved 89% true positive rate and 11 false negative rate with
97% accuracy. Prediction speed and training time was found
to be 12000 observation/seconds and 1.5238 seconds
respectively. This classification model was up to the mark but
accuracy can be further improved. Data can be regularized in
discriminant analysis classifier for the robust classification
model.
Fig. 3. Linear Discriminant Analysis Confusion Matrix.
B. Logistic Regression Classification Model
Fig. 4 represents that Logistic regression classification
model was developed in the MATLAB using the collected
data set. The classification model achieved 96.4% accuracy.
Prediction time and training time was found to be 15000
observation/second and 3.9633 seconds. It took almost double
time to classify the faulty data compared to the linear
discriminant analysis with slight less accuracy.
C. Quadratic Support Vector Machine Classification Model
Fig. 5 illustrates that Quadratic support vector machine
classifier has been applied to the collected data set form the
sea shore of Kund Malir. The model is based on predict SVM
model in which predictors have been defined in the matrix and
then comparison would be performed between the observed
and predicted. The trained Quadratic SVM model may be Fig. 4. Logistic Regression Confusion Matrix.
compact or full. The trained model has been exported for the
testing purpose. yout = predict(QSVMModel,X). Quadratic D. Fine k-Nearest Neighbor Classification Model
support vector machine classifier model achieved 93% true Fig. 6 shows the confusion matrix for the Fine k-nearest
positive rate and 7% False negative rate in the confusion neighbor algorithm. Fine k-nearest neighbor classification
matrix. 96.8% accuracy achieved by Q-SVM in 0.9237 model has been developed by measuring the standardized
seconds of training time. Prediction speed was found to be Euclidian distance.
40000 observations per second.
E=(u −V )B−1(u −V ) (1)
u v u v
355 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
356 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Prediction
Predictive Training
RMSE R2 MSE MAE speed per
Models Time (s)
second
Linear
0.1900 0.82 0.0361 0.113 11000 4.5756
Regression
Interactions
0.1087 0.94 0.0118 0.044 39000 0.8332
Linear
Robust
0.267 0.65 0.0712 0.071 80000 0.898
Linear
Step Wise
0.1118 0.94 0.0125 0.046 70000 3.779
Linear
Linear
Fig. 9. Ensemble Bagged Trees Confusion Matrix. 0.2316 0.73 0.0536 0.127 47000 11
SVM
Quadratic
0.5551 -0.53 0.0308 0.322 73000 13.159
SVM
Gaussian
0.0721 0.97 0.0052 0.045 140000 0.818
SVM
Rational
Quadratic 0.0006 1 3×10-7 9×10-5 22000 1.1555
GPR
Exponential
0.0002 1 4×10-8 3×10-5 35000 2.1752
GPR
Ensemble
Bagged 0.0750 0.97 0.0056 0.0222 34000 2..3892
Trees
Fig. 10. Graphical Analysis of Prediction Model.
A. Graphical Illustration of Linear Regression, Interactions B. Graphical Illustration of Linear SVM, Quadratic SVM,
Linear, Robust Linear and Step Wise linear Gaussian SVM and Rational Quadratic GPR
Fig. 11 demonstrates the graphical illustration of model 1 Fig. 12 demonstrated the graphical illustration of model 5
(Linear Regression), Model 2 (Interactions Linear), Model 3 (Linear SVM), Model 6 (Quadratic SVM), Model 7 (Gaussian
(Robust Linear) and Model 4 (step wise linear). Moreover, SVM) and Model 8 (Quadratic GPR). Moreover, Blue color
Blue color represents the true data and yellow color depicts represents the true data and yellow color depicts the predicted
the predicted data. The graphs have been plotted between data. The graphs have been plotted between number of records
number of records and flood response (hurricane response). and flood response (hurricane response).
357 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
358 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[5] T. Khan, K. Kadir, M. Alcm, Z. Fchiihid and M. S. Mazliham, [13] L. S. Solanki, S. Singh, and D. Singh, "An ANN approach for false
"Geomagnetic field measurement at earth surface: Flash flood alarm detection in microwave breast cancer detection," in 2016 IEEE
forecasting using tesla meter," 2017 International Conference on Congress on Evolutionary Computation (CEC), 2016, pp. 1370-1374.
Engineering Technology and Technopreneurship (ICE2T), Kuala [14] O Daniell, Y Petillot, and S Reed., ―Unsupervised seafloor classification
Lumpur, 2017, pp. 1-4. for automatic target recognition‖. Proc. International Conf. Remote
[6] F. A. Ruslan, A. M. Samad and R. Adnan, "Modelling of flood Sens., (October), 2012.
prediction system using hybrid NNARX and Extended Kalman [15] DP Williams and E Fakiris. Exploiting environmental information for
Filter," 2017 IEEE 13th International Colloquium on Signal Processing improved underwater target classification in sonar imagery. IEEE Trans.
& its Applications (CSPA), Batu Ferringhi, 2017, pp. 149-152. Geosci. Remote Sens., 52(10):6284–6297, 2013.
[7] T. A. Khan, M. Alam, K. Kadir, Z. Shahid and S. M Mazliham, "A [16] C. E. Rasmussen, Evaluation of Gaussian processes and other methods
Novel Approach for the Investigation of Flash Floods using Soil Flux for non-linear regression, University of Toronto, 1999.
and CO2: An Implementation of MLP with Less False Alarm
[17] Suresh, P. V. S., Venkateswara Rao, P. and Deshmukh, S. G., ―A
Rate," 2018 2nd International Conference on Smart Sensors and
Application (ICSSA), Kuching, 2018, pp. 130-134. Genetic Algorithmic Approach for Optimization of Surface Roughness
Prediction Model,‖ International Journal of Machine Tools and
[8] M. Hijji, S. Amin, R. Iqbal and W. Harrop, "A Critical Evaluation of the Manufacture, Vol. 42, 2002, pp. 675-680.
Rational Need for an IT Management System for Flash Flood Events in
[18] Yujun Yang, Jianping Li and Yimei Yang, "The research of the fast
Jeddah, Saudi Arabia," 2013 Sixth International Conference on
Developments in eSystems Engineering, Abu Dhabi, 2013, pp. 209-214. SVM classifier method," 2015 12th International Computer Conference
on Wavelet Active Media Technology and Information Processing
[9] M. Mousa, X. Zhang and C. Claudel, "Flash Flood Detection in Urban (ICCWAMTIP), Chengdu, 2015, pp. 121-124.
Cities Using Ultrasonic and Infrared Sensors," in IEEE Sensors Journal,
vol. 16, no. 19, pp. 7204-7216, Oct.1, 2016. doi:10.1109/JSEN.2016. [19] C. E. Rasmussen, C. K. Williams, Gaussian processes for machine
2592359. learning, Vol. 1, MIT press Cambridge, 2006.
[20] Talha Khan, Muhammad Alam, Faraz Shaikh, Sheroz Khan, Kushsairy
[10] T. A. Khan, M. Alam, Z. Shahid and M. M. Suud, "Prior investigation
Kadir, Zeeshan Shahid, M.S Mazliham & Yahya,‖ Flash floods
for flash floods and hurricanes, concise capsulization of hydrological
prediction using real time data: An implementation of ANN-PSO with
technologies and instrumentation: A survey," 2017 IEEE 3rd
less false alarm‖, 2019 IEEE International Instrumentation &
International Conference on Engineering Technologies and Social
Sciences (ICETSS), Bangkok, 2017, pp. 1-6. Measurement Technology Conference 20-23 May 2019, Grand
Millenium Auckland, New Zealand.
[11] N. Bhardwaj, N. Aggarwal, N. Ahlawat and C. Rana, "Controls and
[21] T. Khan, M. Alam, and M. Mazliham, ―Artificial Intelligence Based
intelligence behind ―NISTARA-2‖—A disaster management machine
Multi-modal Sensing for Flash Flood Investigation‖, jictra, pp. 40-47,
(DMM)," 2014 Innovative Applications of Computational Intelligence
Jun. 2018.
on Power, Energy and Controls with their impact on Humanity
(CIPECH), Ghaziabad, 2014, pp. 34-37. [22] F. Mercier, N. Akrour, L. Barthès, C. Mallet and R. Hallali, "Fine-scale
[12] Kotsiantis S.B., Tsekouras G.E., Pintelas P.E. (2005) Bagging Model evaluation of rainfall from TV-sats: A new method for water monitoring
Trees for Classification Problems. In: Bozanis P., Houstis E.N. (eds) and flash flood prevention," in URSI Radio Science Bulletin, vol. 2017,
no. 360, pp. 80-88, March 2017.
Advances in Informatics. PCI 2005. Lecture Notes in Computer Science,
vol 3746. Springer, Berlin, Heidelberg.
359 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Conceptual modeling is an essential tool in many The UML [8] has been utilized as a graphical notation to
fields of study, including security specification in information construct and visualize security aspects in object-oriented
technology systems. As a model, it restricts access to resources systems. It is currently utilized as a primary notation for
and identifies possible threats to the system. We claim that security and authentication because it provides a spectrum of
current modeling languages (e.g., Unified Modeling Language, notations representing the various aspects of a system. The use
Business Process Model and Notation) lack the notion of of the UML for conceptual modeling requires special care to
genericity, which refers to a limited set of elementary processes. not confuse software features with aspects of the real world
This paper proposes five generic processes for modeling the being modeled [9].
structural behavior of a system: creating, releasing, transferring,
receiving, and processing. The paper demonstrates these BPMN was designed to be used by people without much
processes within the context of public key infrastructure, training in software development. ―UML diagrams look
biometric, and multifactor authentication. The results indicate technical, and in practice, they are much harder for
that the proposed generic processes are sufficient to represent businesspeople to understand than BPMN diagrams‖ [10].
these authentication schemes. BPMN includes a rich set of model constructs for business
process modeling.
Keywords—Security; authentication; conceptual modeling;
diagrammatic representation; generic processes This paper is about conceptual modeling. It is part of a
research project that applies a new modeling language, the
I. INTRODUCTION thing machine (TM), to modeling computer attacks [11]. The
Security is a necessary feature in information technology paper concentrates on using the TM to model authentication.
(IT) systems. Security specification requires identifying risks, The thesis promoted in our research works is that modeling in
access requirements, and recovery strategies, and comprises the abovementioned languages lacks genericity, a notion for
well-developed security mechanism processes [1]. Early-stage representing systems that forms the base for process modeling.
development of security specification assists in lowering the This has caused conceptual vagueness that obstructs the
possibility of security breaches. differentiation of objects. A specific goal of the paper is to
substantiate the viability of the TM by applying it to modeling
Authorization and authentication both play vital roles in the authentication.
configuration of security mechanisms. Authorization is the
process of allowing users to access system objects based on B. Modeling Authentication
their identities. Authentication confirms that the user is who he In the twenty-first century, few matters are more pressing
or she claims to be. than those related to identity authentication. Authentication is a
Conceptual modeling is a description of reality using a mechanism used to make sure that those obtaining session
modeling language to create a more-or-less formalized schema access are who they say they are. To access online systems and
[2]. A conceptual model in the security field restricts access to services, we all face the challenge of proving our identities
the resources and identifies possible threats to the system. In [12].
modeling, notations (diagrams, symbols, or abbreviated In the real world, thousands have found themselves blocked
expressions) are required to specify technical facts and related from opening bank accounts, making payments, or travelling
concepts of systems. They are necessary to articulate complex because of an unfortunate name similarity to those individuals
ideas succinctly and precisely [3]. For a notation to convey or entities on a sanctions list. Hundreds of thousands have been
accurate communication, it must effectively represent the victims of identity fraud, often only learning of the crime when
different aspects of a system and be well understood among they apply for credit and find their credit rating has been
project participants. The historic roots of modeling notations in compromised by fraudulent loans obtained in their names [12].
software engineering can be traced back to structured analysis
and design, which are based on data flow diagrams [3]. In this paper, we focus on individual and entity
authentication for digital interactions. We concentrate on
A. Security Modeling authentication in the context of usability of IT systems in terms
Many languages and mechanisms, such as Business Process of who is using the system, what they are using it for, and the
Model and Notation (BPMN) [4], secure Tropos [5], misuse environment in which they are using it (ISO standard 9241 Part
cases [6], and mal-activity diagrams [7], are used in the field of 11). The ISO 9241 standard for identity authentication is made
security modeling. For space consideration, we focus here on up of three components: what you are (e.g., biometric
the Unified Modeling Language (UML) and BPMN. information), what you have (e.g., having a token), and what
you know (e.g., PINs, passwords).
360 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
361 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
362 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
This is what we mean by generic processes. Even though described, the queue has a rear, which, upon the arrival of the
they are used differently according to the setting, members of customer (2), is retrieved/released (3) and incremented (4).
each generic process seem to be synonymous with respect to Hence:
things. In language, such a phenomenon appears in the case of
the adjectives big, great, and large, which are seemingly If Q is full (the maximum capacity of the queue when
synonymous words but are likely to be used in different ways (rear+1)mod n =front), the system blocks any newly
in different settings [27]. Processes recognized as being of the arriving customers.
same kind of ―meaning‖ in the above sense are said to possess The new rear value is stored (6).
a generic property. Generic processes are conduits through
which various types of processes flow. Accordingly, the customer is assigned a position (given a
number) in the queue and joins the other customers waiting in
III. THING MACHINE MODELING EXAMPLE the queue (8).
Guizzardi and Wagner [2] give an example of a service Whenever the service agent is not busy (9):
queue system in which customers arrive at random times at a
service desk. They have to wait in a queue when the service The first customer in the queue is released to the
desk is busy. Otherwise, when the service desk is not busy, service area (10).
they are immediately served by the clerk. Whenever a service The arrival of the customer to the service area changes
is completed, the next customer from the queue (if any) is its state to busy (11).
served [28].
The customer is then processed (12).
Fig. 7 shows the TM model of the example. The customer
arrives (circle 1) to get into the queue (Q). We assume a The customer is released (13), which triggers the not
circular queue structure stored in Q(0:n - 1) with mod n busy state (14).
operation; rear points to the last item and front is one position
counterclockwise from the first item in Q. As typically The customer leaves the service area (15).
363 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Triggering the not busy state results in taking a new Event 3 (E3): A customer joins the queue.
customer from the queue, as mentioned previously (10), and Event 4 (E4): Top is retrieved and incremented, and the new
also updates the queue data (16). Thus, value is stored.
Event 5 (E5): The queue is full (i.e., new value = max).
The front value is retrieved (17) and decremented (18), Event 6 (E6): The queue is not full.
and the new value is stored (19). Event 7 (E7): A customer joins the queue.
The original front value (before decrementing it) is Event 8 (E8): The service agent is not busy.
checked (20), and if the Q was full, the blockage of Event 9 (E9): A customer moves from the queue to the service.
new customers from entering the queue is lifted (21). Event 10 (E10): The service becomes busy.
Event 11 (E11): The customer leaves the service.
Initially, we assume that the entrance is not blocked, the Event 12 (E12): Top is retrieved and decremented, and the new
queue is empty, and the service is not busy. value is stored.
The dynamic behavior of the system can be developed Event 13 (E13): Top becomes less than max.
based on events. An event in a TM is treated as a Fig. 10 shows the behavior of the system in terms of the
thing/machine—that is, it can be created, processed, released, chronology of its events.
transferred, and received. For example, the event a customer
moves from the queue to the service desk is represented as
shown in Fig. 8. It has two submachines: time and region
where the event takes place. An event also denotes a change.
All stages in the static description of Fig. 7 indicate elementary
changes; however, we are typically interested in larger events
that include several stages, as demonstrated in the event a
customer moves from the queue to the service. Accordingly,
we identify the following events in this example (see Fig. 9):
Event 1 (E1): The service is open.
Event 2 (E2): The service is closed (blocked).
Fig. 8. Event with Region and Time Submachines.
364 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 10. Chronology of Events. Fig. 12. Description of UML use Case Diagram—SA. (Partially Taken from
[29]).
IV. CASE STUDY: MODELING AUTHENTICATION
To apply a TM to modeling authentication, we adopt a
security case study that involves insider attackers as presented
by Nostro et al. [29]. This case study is interesting because it
adopts a modeling approach using UML diagrammatic and
textual use cases in line with the level of modeling applied in
this paper. Additionally, UML use cases give us an opportunity
to contrast use case diagrams with TM diagrams.
The case study includes the taxonomy of users physically
or logically involved within the system and investigates their
roles as potential insiders. The users are system administrator
(SA), system expert, unknown user, domain expert, human
sensor, and operator. Nostro et al. [29] explore only the SA and Fig. 13. Sample Attack Execution Graph. (Partially Redrawn from [29]).
system expert, and we, in this paper, focus on the SA
performing a software update. Fig. 11 shows the use case We claim in this paper that the TM model presents a
related to the SA; the darkened part indicates our region of systematic alternative (one kind of notion) in modeling
emphasis. Fig. 12 shows the textual description of the use case. security. Without loss of generality, we will focus on the
authentication part of Nostro et al. [29] to demonstrate the
Based on such a use case model that ―guides the whole viability of the TM model.
process,‖ Nostro et al. [29] identify and assess insider threats
and develop countermeasures that are oriented toward V. MODELING AUTHENTICATION
prevention, deterrence, or detection. They also use an ad hoc Authentication plays an important role in the security of
attack execution graph called ADVISE (see Fig. 13). computing, hence the existence of several authentication
techniques. An authentication process attempts to verify a
user‘s identity prior to the user‘s access to any resources in
order to protect the system against various attack types. Once
authenticated, the user is permitted to connect with cloud
servers to request services [30-33]. Without loss of generality
and due to space limitations, we will apply the TM model to
only three authentication methods: public key infrastructure
(PKI) authentication, biometric authentication, and multifactor
authentication. As discussed in the case study in Section IV,
the authentication of the SA is a precondition of all four use
cases (system maintenance, data management, profile
management, and crisis management, as represented in
Fig. 11). The login session allows the SA to begin requesting
services from the system. However, no requests from any of
Fig. 11. UML use Case Diagram Involving the SA. (Partially redrawn from these four use cases will be serviced until the SA is
[29]). authenticated by the system.
365 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The first SA role to be investigated is the system o The system uses the SA‘s public key (10) to decrypt
maintenance case. This case is an umbrella to three subcases (11) the incoming encrypted data, producing
involving software updates, installing software, and managing decrypted data (12).
servers.
o The decrypted data are compared (13) to the original
A. Public Key Infrastructure Authentication random data; if they are equivalent, a system
Fig. 14 shows the TM representation of SA roles under the maintenance session is opened for the SA (14).
PKI framework system, whereas Fig. 15 shows the
A selected set of events are described as follows (see Fig.
corresponding dynamic system, assuming the SA is already
15):
certified. The figure comprises two main machines: the SA and
the system (highlighted in yellow). Event 1 (E1): The SA logs into his or her account, and the system
creates a session accordingly.
The SA logs into his or her account (Circle 1 in Event 2 (E2): The SA issues a request to maintain the system.
Fig. 14). Event 3 (E3): The system starts the authentication process by
generating random data and sending it to the SA.
Assuming correct credentials, the system creates (2) a
Event 4 (E4): The SA processes the random data using his or her
session.
private key and sends the encrypted data to the system.
The SA issues a request (3) for system maintenance, Event 5 (E5): The system uses the SA‘s public key to decrypt the
such as a software update. incoming encrypted data, producing a decrypted dataset.
Event 6 (E6): The original random data are transferred to the
Upon receiving the request, the system performs the comparison module.
authentication process (4) [34] as follows: Event 7 (E7): The decrypted data are compared to the original
random data.
o The system generates random data (5) using the Event 8 (E8): If the data are equivalent, a system maintenance session
SA‘s public key and sends it to him or her (6). is opened for the SA.
o The SA processes (7) the random data using his or Fig. 16 shows the chronology of these events that model
her private key (8) and sends its encrypted version the behavior of the PKI-based authentication system.
to the system (9).
Fig. 14. TM Representation of UML use Case Involving the SA in PKI Authentication.
366 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Biometric Authentication o The system then extracts (12) the scanned trait for
comparison purposes.
Fig. 17 and 18 show the static and dynamic TM
representations of the SA‘s roles under a physical biometric o The originally encrypted trait is decrypted (13) and
authentication system. A typical physical biometric system compared with the trait extracted from the scanning
carries out authentication in two stages—the enrollment stage hardware (14). If they are equivalent, a system
and the verification stage. maintenance session is opened to the SA (15).
Fig. 17 comprises two main machines: the SA and the Fig. 18 shows the dynamic description of the model. A
system (highlighted in yellow). selected set of events is described as follows:
Initially, in the enrollment stage, the SA requests (1) Event 1 (E1): The SA requests the biometric trait desired for the
the biometric trait desired, such as a face or fingerprint. enrollment stage, and the system requests the SA to present the chosen
biometric trait.
In response, the system requests (2) the SA to present Event 2 (E2): The SA presents the trait to the scanning hardware for
his or her chosen biometric trait. extraction.
The SA then presents (3) the trait to the scanning Event 3 (E3): The extracted data are then encrypted and stored.
hardware. Event 4 (E4): The SA logs into his or her account, and the system
creates a session accordingly.
The system then extracts (4) the scanned trait for Event 5 (E5): The SA issues a request for system maintenance.
encryption and storage (5). Event 6 (E6): The system starts the authentication process by
requesting the SA to present the chosen biometric trait.
To initiate an interaction with the system, the SA logs Event 7 (E7): The SA presents the trait to the scanning hardware for
into his or her account (6). With the correct credentials, extraction.
the system creates (7) a session. Event 8 (E8): The system decrypts the originally encrypted trait.
Event 9 (E9): The extracted trait is compared to the decrypted data.
The SA issues a request (8) to maintain the system Event 10 (E10): If the data are equivalent, a system maintenance
(e.g., software update). session is opened to the SA.
367 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 17. TM Representation of UML use Case Involving the SA in Physical Biometric Authentication.
368 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
369 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[20] S. Al-Fedaghi and E. Haidar, ―Programming is diagramming is [29] N. Nostro, A. Ceccarelli, A. Bondavalli, and F. Brancati, ―Insider threat
programming,‖ 3rd International Conference on Computer, Software assessment: A model-based methodology,‖ Op. Syst. Rev., vol. 48, no.
and Modeling, Barcelona, Spain, July 14–16, 2019. 2, pp. 3–12, December 2014.
[21] M. Heidegger, ―The thing,‖ in Poetry, Language, Thought, A. [30] S. M. Dejamfar and S. Najafzadeh, ―Authentication techniques in cloud
Hofstadter, Trans. New York: Harper & Row, 1975, pp. 161–184. computing: A review,‖ Int. J. Adv. Res. Comput. Sci. Softw. Eng., vol.
[22] K. Riemer, R. B. Johnston, D. Hovorka, and M. Indulska, ―Challenging 7, no. 1, pp. 95–99, January 2017.
the philosophical foundations of modeling organizational reality: The [31] A. Banerjee and M. Hasan, Token-Based Authentication Techniques on
case of process modeling,‖ International Conf. on Information Systems, Open Source Cloud Platforms, Systems and Telematics, Vol. 16, No.
Milan, Italy, 2013. http://aisel.aisnet.org/icis2013/proceedings/ 47, pp. 9-29, October-December, 2018.
BreakthroughIdeas/4/. [32] M. Qasaimeh, R. Turab, R. S. Al-Qassas, Authentication techniques in
[23] S. Al-Fedaghi, ―Five generic processes for behaviour description in smart grid: a systematic review, TELKOMNIKA, Vol.17, No.3,
software engineering,‖ Int. J. Comp. Sci. Inf. Secur., vol. 17, no. 7, July pp.1584-1594, June 2019.
2019. [33] A. Agarkar and H. Agrawal, A review and vision on authentication and
[24] S. Al-Fedaghi, ―Toward maximum grip process modeling in software privacy preservation schemes in smart grid network, Security and
engineering,‖ Int. J. Comput. Sci. Inf. Secur., vol. 17, no. 6, June 2019. Privacy, Vol. 2, No. 2, pp. 1-18,March/April 2019.
[25] L. W. Howe, ―Heidegger‘s discussion of ‗the Thing‘: A theme for deep [34] M. Furuhed (2018). Public key infrastructure (PKI) explained in 4
ecology,‖ Between Species, vol. 9, no. 2, art. 11, 1993. minutes, Nexusgroup.com, accessed 5/8/2019. https://www.
doi:10.15368/bts.1993v9n2.9. nexusgroup.com/blog/crash-course-pki.
[26] L. R. Bryant, ―Towards a machine-oriented aesthetics: On the power of [35] W. Yang, S. Wang, J. Hu, G. Zheng, and C. Valli, ―Security and
art,‖ paper presented at The Matter of Contradiction Conference, accuracy of fingerprint-based biometrics: A review,‖ Symmetry, vol. 11,
Limousin, France, 2012. no. 2, art. 141, January 2019. https://www.mdpi.com/2073-
[27] P. Byrd, Generic Meaning, accessed 5/8/2019. 8994/11/2/141.
http://www2.gsu.edu/~eslhpb/grammar/lecture_5/generic.html [36] K. Garska (2018). Two-Factor Authentication (2FA) Explained: Email
[28] G. Guizzardi and G. Wagner, ―Tutorial: Conceptual simulation and SMS OTPs, Identity Automation Site, September 27, 2018.
modeling with onto-UML,‖ Proceedings of the 2012 Winter Simulation https://blog.identityautomation.com/two-factor-authentication-2fa-
Conference, Berlin, Germany, December 9–12, 2012. explained-email-and-sms-otps.
370 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This study aims to create a network security system network, it is done so to avoid incoming intrusion flows from
that can mitigate attacks carried out by internal users and to the outside [11]. Thus, concerns that attacks can still arise from
reduce attacks from internal networks. Further, a network inside intruders to network before reaching IPS still exists in
security system is expected to be able to overcome the difficulty the network. Therefore, a network security is needed from the
of mitigating attacks carried out by internal users and to improve lowest level of the internal network as well.
network security. The method used is to integrate the ability of
Network Access Control (NAC) and the Intrusion Prevention Network Access Control (NAC) is an approach designed to
System (IPS) that have been designed and implemented in this increase network security by controlling the access and the
study, then an analysis is performed to compare the results of resources for legitimate users [12]. NAC not only allows
tests that have been carried out using only the NAC with the network access requested by the user, but also provide specific
results using integration of NAC capabilities and IPS. The results access based on the user's identity [13]. One of the threats to
obtained from the tests that have been carried out, namely, the enterprise networks is the personal devices of employees and
security system by using the integration of NAC and IPS guests that do not have anti-virus, patches or host intrusion
capabilities is better than using only the NAC. prevention system in place. An NAC solution can protect a
network from such end devices and detect and rectify these
Keywords—Attack; integration; Intrusion Prevention System problems [14]. NAC function has certain weaknesses, in
(IPS); mitigation; Network Access Control (NAC); network security
particular it is unable to detect and stop users that have
I. INTRODUCTION legitimate network access form carrying out intentional or
unintentional attacks from within. Example of intentional
For decades, technology plays an important role in most attack is when an internal user has a desire to destroy the
activities. Most organizations use technology to support their internal system due to personal problems, whereas
business processes. Nowadays, internet is used for almost all unintentional attacks can happened through downloading files
activities especially business activities. Thus, network or applications that contain malware or viruses. This condition
infrastructure plays a vital function in an organization. Most can happen because NAC does not have the ability to detect
organizations are connected to the internet to make all attacks like IPS.
information easily accessed from anywhere and anytime.
Network can also be considered as a major risk for an Based on the weaknesses of the NAC, there is a need to
organization. Today's advancement of IT technology bring to improve network access security from within. In this research,
the surface the issue of security. Thus, it is important to secure a solution is proposed to improve network security from
the network infrastructure [1] [2]. In the operation of network internal sides of the network by integrating NAC and IPS
can be compromised by any vulnerability in their functionality capabilities. The benefits obtained from this solution can
to attack the networks. Some mechanisms are widely used to minimize the threat of attacks on the network.
secure the network, namely Intrusion Detection System (IDS)
that has the ability to detect malicious and unauthorized II. REVIEW OF RELATED LITERATURE
activities and Intrusion Prevention System (IPS) that has the A. NAC
ability to make an action for detected intrusion [3] [4] [5]. The
NAC systems combine endpoint security solutions to grant
purpose of using IDPS is to monitor and protect attacks from
access control and enforce security rules or policies to every
intruder who want to enter the system, and then give a report to
device connected to the network. The NAC policy is able to
the network administrator if there are attacks that occur in the
identify endpoints that are connected to the network. This
network environment [6] [7]. So, using IDPS can help to detect
policy is carried out to restrict access of devices that do not
and carry out security against intrusions that occur on the
comply with predetermined network access rules [15]. NAC
network.
also provides security and control for those who have access to
Attack threats can be caused by either outsiders or insiders networks and resources within the network. Basically, NAC
in an organization. Insider attacks are malicious attacks carried performs posture, quarantine, and remediation checks involved
out on networks or computer systems with authorized/official in requests for network access by users. If the user does not
system access [8] [9]. Insider attack is one of the most difficult have the appropriate posture in his computer such as the latest
threats to be detected because (IDS) is built to defend against OS/security patch or the most updated antivirus, then the user
outside attacks [10]. Generally, IPS is placed in the edge of a will not be allowed to enter the network, but the user will be
371 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
quarantined by being separated into different networks or that cannot be detected by a firewall and NAC can be detected
VLANs until the user performs a remediation process to meet properly and therefore increase network security.
the requirements needed for entry into the network [16] [15].
III. METHODOLOGY
Some reasons for using the NAC solution are: to identify
and authenticate users and endpoints, to limit user access to the A. System Design
network, to limit access based on the endpoint security posture, As summarized in Table I, IPS used in this study is
and to remediate an endpoint if the endpoint does not have a products from Cisco, Cisco FirePower 8250 with OS 6.2.3
posture that complies with the provisions [17]. Another reason series. The IPS has been connected between a firewall device
for implementing the NAC is due to the threat that comes from and core switch. Using the NAC system from Cisco requires a
using your own device (BYOD) approach. With many users Cisco NAC device called the Cisco Identity Service Engine
using their own devices to work and use them for work (ISE). Cisco ISE is a Cisco appliance used for NAC systems.
purposes, NAC is increasingly needed because many security The ISE will be linked and integrated with existing network
threats might occur due to devices that do not have enterprise- infrastructure devices such as switches and radius servers to
level device security standards such as patch OS and antivirus. authenticate. Physically, the Cisco ISE will be connected to the
Comparative studies of existing NAC systems has server-farm switch. This is done so that Cisco ISE can be
concluded that the NAC solutions from Cisco, Trustwave, and integrated with all segments in the network infrastructure. The
Forescout can be implemented in accordance with the existing hardware of Cisco ISE used is appliance with SNS 3495 type
network infrastructure so that it can produce maximum profits and the OS version of Cisco ISE used is ISE 2.3. The NAC will
where NAC can limit the access of devices and users that are be able to communicate with IPS to carry out the expected
defined based on existing roles and ensure network access integration in accordance with the objectives of this study. The
obtained according to what is needed [17]. The main benefit of access switch that directly connected to the user’s PC uses
NAC systems is to prevent potentially malicious or infected Cisco Catalyst 2960X with 15.2(2)E7 IOS version. The
devices from entering the network in order to keep the network computer used as an attacker and the target is HP ProOne 600
clean [18]. So that network security can be increased from the using windows 10.
user level by using NAC. Fig. 1 below proposes a new topology using Cisco
FirePower and Cisco ISE connected to the network. In this
B. IDS and IPS
study, the device was integrated with existing infrastructure.
There are a couple of widely used mechanisms to secure The integration carried out in this study is physical and logical
the network, namely Intrusion Detection System (IDS) and connection where the Cisco ISE NAC and IPS Cisco
Intrusion Prevention System (IPS). Intrusion detection is the FirePower must be able to connect with existing infrastructure
process of monitoring events on a network or computer system devices, change server farm connections from core switch
and analyzing them for possible threat incidents and violations through IPS, configure to integrate between the Cisco ISE
of standard computer security practices, usage policies, or NAC, IPS Cisco FirePower and existing infrastructure devices,
security policies [3]. IDS is a hardware component or software as well as making policies and rules on the Cisco ISE NAC and
that automates the intrusion detection process. It monitors IPS Cisco FirePower to achieve the objectives in this study
events that occur on network and computer systems and which are to create a network security system that can mitigate
responds to alert with an indication of potential network internal users who carry out attacks and to reduce attacks from
security policy violations [19] [20]. IPS is a network device or internal networks by using NAC and IPS system integration.
software that identify and block network threats by assessing
each and every packet based on the network protocols in the B. Implementation and Testing
network layer, tracking each session. Intrusion Prevention In this study, tests were carried out to prove the solution
System is a defense mechanisms designed to detect malicious given to address the existing problems. These tests were
packets within network traffic and stop intrusions, blocking the carried out by using a system and infrastructure design that has
aberrant traffic automatically before it does any [3] [21]. IPS is been integrated with the NAC and IPS systems. The tests took
an improvement from IDS because it does not only have the place by trying to connect an internal user PC to the internal
ability to detect intrusion, but also can take action against network with the Windows 10 operating system which acted as
intrusion or potential malicious network activity [22] [23]. an attacker, placing the target server connected to the firewall
There are several approaches that have been carried out by using workstations with vulnerable OS installed. Then, worked
previous researchers. The objective to be achieved in their on the IPS and NAC configuration so that the two systems can
research is to evaluate and analyze the performance of NGIPS communicate and integrate in order to achieve the objectives of
in securing networks through penetration tests using HTTP this study.
ports, so that the inspection and protection performance of TABLE. I. SYSTEM SPECIFICATIONS
NGIPS is known. The benefit of this research is that it can be a
point of reference in improving network security using the Device Vendor OS Version
NGIPS method and to obtain optimal mechanism for PC HP ProOne 600 Windows 10
implementing NGIPS. Based on the results of these penetration Access Switch Cisco Catalyst 2960X 15.2(2)E7
tests, it proves that NGIPS can save attacks that exploit
NAC Cisco ISE SNS 3495 2.3
vulnerabilities from HTTP ports [24]. By using IPS, attacks
IPS Cisco FirePower 8250 6.2.3
372 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
373 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
374 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Result
No User Traffic Target Expectation
NAC NAC & IPS (Proposed solution)
1 Compliant HTTP Vulnerable OS Allow Allowed Allowed
2 Compliant SSH Vulnerable OS Allow Allowed Allowed
3 Compliant SQL Injection Attack Vulnerable OS Block Allowed Blocked
4 Compliant OS Bash Injection Attack Vulnerable OS Block Allowed Blocked
5 Noncompliant HTTP Vulnerable OS Block Blocked Blocked
[9] M. Warren, "Modern IP theft and the insider threat," Computer Fraud &
V. CONCLUSION AND FUTURE WORK Security, no. 6, pp. 5-10, 2015.
Based on the test results performed in this study, the [10] F. Y. Leu, K. L. Tsai, Y. T. Hsiao and C. T. Yang, "An internal intrusion
proposed solution is that by integrating the NAC system with detection and protection system by using data mining and forensic
techniques," IEEE Systems Journal, pp. 1-12, 2015.
the IPS can mitigate attacks from internal users on internal
[11] R. S. Silva and E. L. C. Macedo, "A cooperative approach for a global
networks and attacks from internal networks. That network intrusion detection system for internet service providers," Cyber Security
security with the integration of the NAC systems with the IPS in Networking Conference, vol. 1, pp. 1-8, 2017.
can be increased as compared to the use of the NAC alone. [12] J. F. Matthews, "Challenges to implementing network access control,"
However, this study still has many limitations, particularly on SANS Institute InfoSec Reading Room, p. 2, 2017.
the types of attacks tested. There are so many different types of [13] M. Roopesh, G. Reethika, B. V. Srinath and A. Sarumathi, "Network
attacks on the internet. Therefore, in the future it is access control," International Journal on Computer Science and
recommended to increase the types of attacks carried out in Engineering (IJCSE). Vol. 9, pp. 338-343, 2017.
similar tests and do more detail experiment to compare the [14] M. S. Inamdar and A. Tekeoglu, "Security analysis of open source
application on the internal network with the application on the network access control in virtual networks," International Conference on
Advanced Information Networking and Applications Workshops, vol. 32,
external network in order to achieve a more comprehensive pp. 475-480, 2018.
result. [15] M. A. Muhammad and A. Ayesh, "A behaviour profiling based technique
REFERENCES for network access control systems," International Journal of Cyber-
Security and Digital Forensics (IJCSDF), vol. 8, no. 1, pp. 23-30, 2019.
[1] W. Bul'ajoul, A. James and M. Pannu, "Improving network intrusion
detection system performance through quality of service configuration [16] A. Sood, "Network access control," Rivier Academic Journal, vol. 3, pp.
and parallel technology," Journal of Computer and System Sciences, vol. 1-12, 2007.
81, no. 6, p. 981–999, 2015. [17] T. J. Dildy, "Network access control-has it evolved enough for
[2] W. Bul'ajoul, A. James and S. Shaikh, "A New Architecture for Network enterprises?," ISACA Journal Vol. 4, pp. 1-5, 2016.
Intrusion," IEEE Access, vol. 7, pp. 18558-18573, 2019. [18] K. O. Detken, M. Jahnke, C. Kleiner and M. Rohde, "Combining network
[3] H. A. Razzak, A. Karim, S. S. Handa and M. V. Ramana Murthy, "A access control (nac) and siem functionality based on open source," in
methodical approach to implement intrusion detection system in hybrid IEEE International Conference on Intelligent Data Acquisition and
network," International Journal of Engineering Science and Computing, Advanced Computing Systems: Technology and Applications, Bucharest,
vol. 7, no. 3, pp. 4817-4820, 2017. 2017.
[4] G. Ahmed, M. N. A. Khan and M. Shamraiz, "A linux-based IDPS," [19] R. R. Chaudhari and S. P. Patil, "Intrusion detection system:
Computer Fraud & Security, pp. 13-18, 2015. classification, techniques and datasets to implement," International
Research Journal of Engineering and Technology, vol. 4, no. 2, pp. 1860-
[5] [5] S. P. Anilbhai and C. Parekh, "Intrusion detection and prevention
1866, 2017.
system for IoT," International Journal of Scientific Research in Computer
Science, Engineering and Information Technology, vol. 2, no. 6, pp. 771- [20] V. Mahajan and S. K. Peddoju, "Deployment of intrusion detection
776, 2017. system in cloud: a performance-based study," IEEE Computer Society,
pp. 1103-1108, 2017.
[6] S. Khadafi, B. D. Meilani and S. Arifin, "Sistem keamanan open cloud
computing menggunakan ids (intrusion detection system) dan ips [21] R. Jamar, A. Sogani, S. Mudgal, Y. Bhadra and P. Churi, "E-shield:
(intrusion prevention system)," Jurnal IPTEK, vol. 21, no. 2, pp. 67-76, detection and prevention of website," IEEE International Conference On
2017. Recent Trends in Electronics Information & Communication Technology,
vol. 2, pp. 706-710, 2017.
[7] F. Arsin, M. Yamin and L. Surimi, "Implementasi security system
menggunakan metode IDPS (intrusion detection and prevention system) [22] B. Y. Choi and D. G. Allison, "Intrusion prevention and detection in
dengan layanan realtime notification," semanTIK, vol. 3, no. 2, pp. 39-48, small to medium-sized enterprises," in SAIS, 2017.
2017. [23] P. Rengaraju, V. R. Ramanan and C.-H. Lung, "Detection and prevention
[8] A. Borkar, A. Donode and A. Kumari, "A survey on intrusion detection of DoS attacks in software-defined cloud networks," IEEE Conference on
system (IDS) and internal intrusion detection and protection system Dependable and Secure Computing, pp. 217-223, 2017.
(IIDPS)," in International Conference on Inventive Computing and [24] G. Duppa and N. Surantha, "Evaluation of network security based on next
Informatics, 2017. generation intrusion prevention system," Telkomnika, vol. 17, no. 1, pp.
39-48, 2019.
375 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This paper presents a new algorithm called CBRm impute time series for short-gaps, between 1 and 2 consecutive
for univariate time series imputation of medium-gaps inspired by NAs. Something similar happens with the LANN and LANN+
the algorithm called Case Based Reasoning Imputation (CBRi) algorithms that were also designed for short-gaps.
for short-gaps. The performance of the proposed algorithm is
analyzed in meteorological time series corresponding to The CBRm imputation process is shown in Fig. 2. As can
maximum temperatures; also it was compared with several be seen when a value between prior and next is calculated, it is
similar techniques. Although the algorithm failed to overcome in not assigned immediately after prior, but is assigned to the
some cases to other proposals regarding precision, the results center of the NA series by doubling in the case that the total of
achieved are encouraging considering that some weaknesses of NAs is an even number.
other proposals with which it was compared were outperformed.
Additionally, this work also presents the results achieved
Keywords—Case Based Reasoning; CBR; CBRm; univariate by the algorithms called Local Average Nearest Neighbors
time series imputation; medium-gaps LANN [3] and LANN+ [3] in medium-gaps imputation
processes. So, a small adaptation for these algorithms was
I. INTRODUCTION done, specifically in the part corresponding to the
Time series data exist in nearly every scientific field, determination of the prior and next values.
where data are measured, recorded and monitored, so it is The present work has been organized as follows: in the
understandable that missing values may occur [1]. The second section, a brief description of the work related to
imputation or completeness of missing values in time series is univariate time series imputation is shown. The third section
a very important task, since if it is not performed it is very shows the theoretical bases necessary for a better
complicated or impossible to be able to successfully carry out a understanding of the content of the work. The fourth section
prediction or forecasting process. describes the proposed algorithm and its implementation. The
In the research field of imputation, univariate time series fifth section describes the results achieved, which are
are a special challenge, most of the standard algorithms rely on compared with different univariate time series imputation
inter-attribute correlations to estimate values for the missing techniques. The sixth section shows the conclusions reached in
data [2]. In the univariate case no additional attributes can be the present work and finally in the seventh section, it is
employed directly, so effective univariate algorithms instead indicated, the works that can be carried out based on the results
need to make use of the time series characteristics. of the work presented.
376 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
377 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Missing Data the existing problems and their solutions from the case base are
Depending on what causes missing data, the gaps will have retrieved and re-used.
a certain distribution. Understanding this distribution may be The second principle is that the type of problems which an
helpful in two ways [16]. First, this knowledge can be used to agent faces tends to repeat [18]. Thus, there is similarity
select the most appropriate imputation algorithm to complete between past and current problems or current and future
the NA values. Secondly, this knowledge can help design an problems. Therefore, it is worth to remember and reuse. This
imputation model, which allows the elimination of the NA leads to construction of the case base which contains
values from a set of test data. This model will help generate the completely resolved problems and their respective solutions.
NA values where the true values are known. Therefore, the
quality of the model can be tested through different regression The complete Case Based Reasoning process is shown in
metrics such as RMSE, MAPE, etc. Fig. 3.
Mechanisms of missing data can be classified into three
categories: Missing Completely at Random (MCAR), Missing
at Random (MAR) and Not Missing at Random (NMAR). The
process of completing NA-gaps in time series is sometimes
complicated, since the underlying mechanisms are unknown
[16]. The diagnosis of MAR and NMAR requires a manual
analysis of data patterns and the application of domain
knowledge, while MCAR can be tested with the t-test or
Little’s test [17].
C. Univariate Time Series
This term refers to a time series that consists of single
observations recorded sequentially over successive time
periods. Although a univariate time series is usually considered
as one column of observations, time is in fact an implicit
variable [16].
D. Univariate Imputation Methods
Techniques capable of doing imputation for univariate time
series can be roughly divided into three categories [16]:
Univariate algorithms. These algorithms work with
univariate inputs and commonly do not employ the
time series features. Some of them are: mean, mode,
median, random simple, last observed carried forward,
etc.
Fig. 3. CBR Process.
Univariate time series algorithms. Most of these
algorithms are developed in section II, and some of IV. CBRM
them are: Missing Value Imputation by Weighted
Moving Average [3] (SMA, LWMA and EWMA), CBRm is inspired by the CBRi “unpublished” [4] algorithm
Kalman, ARIMA, ARIMA-Kalman, Local Average of that was designed for short-gaps imputation processes and that
Nearest Neighbors [3] (LANN y LANN+), and Case when applied to medium-gaps imputation processes can
Based Reasoning Imputation (CBRi) among others no present problems of bias towards the prior value. Taking this
cited in this work. assessment into consideration, CBRm begins the imputation
from the middle of the series of consecutive NAs as shown in
Multivariate algorithms on lagged data. Commonly, Fig. 2.
multivariate algorithms cannot be used for univariate
time series. However, using lags and leads it is possible Fig. 4 shows the proposed CBR system within which
to apply multivariate time series algorithms to a implements the CBRm algorithm. Implementation process for
univariate time series and thus take advantage of CBRm is quite similar to CBRi “unpublished” [4], below are
features offered by multivariate algorithms. the required steps to implement it.
378 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Case Base Implementation The algorithm in Javascript language to build the case base
An algorithm was implemented to build the case base. The is shown in Table II “unpublished” [4]. This algorithm aims to
case base matrix consists of something similar to what is create the matrix or case base (Q). It receives as arguments the
shown in Table I. empty Q matrix and a temperature vector, and returns as a
result the matrix of cases Q.
… 17.2 17.3 17.4 17.5 17.6 17.7 17.8 17.9 18.0 18.1 18.2 18.3 18.4 …
16.8*16.4*17.2 16.6*17*16.6
… 17.4*17.4 17.4*17.6*19.4 16.6*18.6 18*18.2*17.6 17*18.8
*17.4*17.5*17 *17.2
17.2
15.8*17*17.2* 17.8*17.2*17
17.3 17.6*17.1 17.2 19.2 17.6*19.4 18.2*17.6 18
18*17.6*17.1 .6
17.4 18.8 17.6
18.2*17.4*17 18*17.8*1 16.8*19*19*17. 17.8*17*17.8*1
17.5 18*18.8*18.4 16.6*18
.8*17.8*17.8 7.2 8*19.1 8.2
17.6 18
17.4*18.8* 18.6*17.4*17.6
18*18.2*18.4
17.7 15*17.4*17.8 18*17.9 17.4*17.8* *17.6*17.6*17. 17.8 17.4*17.6 17.4*18.8
*18*18.8*18
17.5 8*17.2
17.8 18.6
17.8*19.2*18
17.2*18.2*17 .4*18.4*18.4 18.2*18.2*1
17.9 16.8*18.2*17.2 17.8 17.8 17.6*18.2*17.8 18.2*18.2
.8 *18.2*18.4*1 8.8*18.7
8.1
18.0 18.2
17.8*17.6*18.8 17.8*18.6*1
17.4*18.2* 16.4*18*18.6*1
18.1 18*17.4 17*17.8 *17.8*17.8*18. 18*18 9.4*19.4*18.
18.4*17.2 8*18*18.2*18.4
2*18.2 8*18*
18.2
18.6*19.2*18.2 17.6*18*19.8
18.3 18.8 18.8*18.6 17.4 18*17.8 17.6*18.2
*18.2 *18.6*18.2
…
As it shows in Fig. 4, for the first block of code that
TABLE. II. ALGORITHM TO BUILD THE CASE BASE (Q)
corresponds to the determination of the prior and next values
that are required by the getMoreSimilar() function to extract
function fillMatrix(Q,temv)
the most similar case from the case base; these values are
{ nQ=Q.length;
for(i=0; i<nQ; i++)
determined through the code between line 4 and line 18 using
{ prior=temv[i]; for this task the array of positions of the NA values.
for(j=0; j<nQ; j++)
{ next=temv[j];
res=look4cases(prior,next);
if(res!="")
Q[i][j]=res;
}
}
return Q;
}
C. CBRm Implementation
According to Fig. 4, four blocks of code can be seen in the
CBRm algorithm, and their detail can be seen in the code
shown in Table III. The CBRm algorithm receives as inputs the
time series with NA values and an array with the positions of
each NA value.
379 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. III. CBRM ALGORITHM In the last block of code from line 27 to line 40, the NA
value is filled with the value calculated according to the
1. function CBRm(tsna,pos) imputation process shown in Fig. 2. The process is repeated
2. { npos=pos.length; until the time series does not contain NAs values.
3. while(npos>0)
4. { nna=0; V. RESULTS AND DISCUSSION
5. ini1=pos[0];
6. fin1=pos[0]; In this section, the performance of the proposed algorithm
7. pini=0; CBRm is compared with different techniques described in
8. pfin=pini; Section II, the comparative results are shown below.
9. prior=parseFloat(tsna[pos[0]-1]);
10. nav=tsna[ini1]; According to Table IV, for a 15-days maximum
11. while(nav=="NA") temperature time series with 73.33% of NA values, the best
12. { nna++; techniques were LWMA and EWMA in the first place (RMSE
13. fin1++; 0.6941); second is LANN+ (RMSE 0.7077); and thirdly very
14. pfin++; close to the previous one is CBRm (RMSE 0.7083). For a
15. nav=tsna[fin1]; percentage of 60% of NAs, the best technique was LANN+
16. } (RMSE 0.6616); secondly LANN (RMSE 0.7187); and thirdly
17. next=parseFloat(nav); CBRi (RMSE 0.7461). For a percentage of 46.67% of NAs, the
18. fin1--;
best technique was CBRi (RMSE 0.4140); second is LANN
19. data=getMoreSimilar(prior,next);
20. dat=data.split("*");
(RMSE 0.4423); and finally, thirdly EWMA (RMSE 0.4780).
21. ndat=dat.length; It is important to highlight that for the first two cases,
22. s=0; ARIMA Kalman produced no results.
23. for(k=0;k<ndat;k++)
24. s+=parseFloat(dat[k]); A graphical comparison of CBRm RMSE with other
25. NA=(prior+(s/ndat)+next)/3; techniques can be seen in Fig. 5.
26. sNA=NA.toFixed(1);
27. rna=nna%2; TABLE. IV. COMPARISON WITH OTHER UNIVARIATE IMPUTATION
28. pna=Math.floor((ini1+fin1)/2); TECHNIQUES (15 DAYS)
29. del=Math.floor((pini+pfin)/2);
30. if(rna==0) RMSE (NAs RMSE (NAs RMSE (NAs
Technique
73.33%) 60%) 46.67%)
31. { m1=pna;
32. m2=pna+1; CBRm 0.7083 0.8164 0.6152
33. tsna[m1]=smed; CBRi 0.8575 0.7461 0.4140
34. tsna[m2]=smed; LANN 0.8050 0.7187 0.4423
35. pos.splice(del-1,2); LANN+ 0.7077 0.6616 0.6175
36. } Hotdeck 0.9534 0.9189 1.0823
37. else
SMA 0.7323 0.8432 0.4928
38. { tsna[pna]=sNA;
39. pos.splice(del,1); LWMA 0.6941 0.8096 0.5209
40. } EWMA 0.6941 0.7958 0.4780
41. npos=pos.length; ARIMA Kalman NA NA 0.5976
42. }
43. return tsna;
44. }
1
In the second block of code (line 19) the getMoreSimilar () 0.8575 0.8164
function is called, this function implements a similarity search 0.8
in the base of cases (Q) using the prior and next values
0.7083
determined in the previous code block, returning a string 0.7461
RMSE
0.6
0.6152
containing the values that will be used in the next code block.
The getMoreSimilar() function implements Euclidean Distance
0.4
according to equation (1) to determine the similarity between
0.414
two points.
0.2
√( ) ( ) (1)
In the third block of code between lines 20 and 26, the 0
73.33% 60% 46.67%
string returned by the getMoreSimilar () function is used and
the NA value is calculated according to equation (2). NA Percentage
CBRm CBRi LANN LANN+
∑ ( ) Hotdeck SMA LWMA EWMA
( )
(2)
Fig. 5. Comparison with other Techniques (15 Days).
380 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
According to Table V, for 90-day time series with a According to Table VI, the proposed CBRm outperformed
percentage of 80% of NA values, the best technique was the multivariate imputation algorithms KNN and MICE.
CBRm (RMSE 0.6844); second is LWMA (RMSE 0.7673);
and thirdly EWMA (RMSE 0.7682). For a percentage of CBRi “unpublished” [4] despite the weaknesses mentioned,
65.55% of NAs, the best technique was SMA (RMSE 0.7035), of the six problems proposed, as CBRm in two of them was
followed by LWMA (RMSE 0.7083), and thirdly LANN+ among the best: in one of them it reached first place and in
(RMSE 0.7339). For a percentage of NAs of 54.44%, the best another it achieved third place.
technique was LWMA (RMSE 0.8106), followed by SMA LANN [3] for medium-gaps from 6 problems, in 2 of them
(RMSE 0.8403), and thirdly EWMA (RMSE 0.8535) he was among the best, it got second place twice. LANN+ [3]
A graphical comparison of CBRm RMSE with other was a little better tan LANN, in 3 cases it was among the best
techniques can be seen in Fig. 6. getting the first, second and third place.
0.8112
0.8
0.8905 REFERENCES
0.6
0.6844 0.805 [1] Rantou, “Missing Data in Time Series and Imputation Methods,”
University of the Aegean, Samos, 2017.
0.4
[2] S. Moritz, A.Sardá, T. Bartz-Beielstein, M. Zaeffer, J, Stork,
0.2 “Comparison of different methods for univariate time series imputation in
R,” arxiv.org, 2015.
0
80.00% 65.55% 54.44% [3] A. Flores, H. Tito, C. Silva, “Local average of nearest neighbors:
Univariate time series imputation,” International Journal of Advanced
NA Percentage Computer Science and Applications, vol. 10, nº 8, 2019.
CBRm CBRi LANN LANN+
[4] A. Flores, H. Tito, C. Silva, “CBRi: A Case Based Reasoning-Inspired
Hotdeck SMA LWMA EWMA
Approach for Univariate Time Series Imputation. Unpublished,” de IEEE
Latin American Conference on Computational Intelligence, Guayaquil,
Fig. 6. Comparison with other Techniques (90 Days). Ecuador, 2019.
[5] N. Bokde, M. Beck, F. Martinez, K. Kulat, “A novel imputation
TABLE. VI. COMPARING WITH MICE AND KNN (90 DAYS) methodology for time series based on pattern sequence forecasting,”
Pattern Recognition Letters, 2018.
RMSE RMSE RMSE
Technique
(NAs 80%) (NAs 65.55%) (NAs 54.44%) [6] A. Zeileis, G. Grothendieck, “zoo: S3 infrastructure for regular and
irregular time series,” Journal of Statistical Software, vol. 14, nº 6, 2005.
CBRm 0.6844 0.8050 0.8968
[7] K. Kaiser, O. Affuso, T, Beasley, D. Allison, “Getting carried away: A
MICE 1.4063 1.3900 1.4714 note showing baseline observation carried forward (BOCF) results can be
KNN 1.0807 1.0751 1.2388 calculated from published complete-cases results,” PMC US National
Library of Medicine, 2012.
381 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[8] A. Kowarick, M. Templ, “Imputation with the R package VIM,” Journa [14] G. Chang, T. Ge, “Comparison of missing data imputation methods for
of Statistical Software, vol. 74, nº 7, 2016. traffic flow,” de International Conference of Transportation, Mechanical,
[9] T. Aljuaid, S. Sasi, “Proper imputation techniques for missing values in and Electrical Engineering (TMEE), Chanchung, China, 2011.
data sets,” de International Conference on Data Science and Engineering [15] B. Sun, L. Ma, W. Cheng, “An improved k-nearest neighbours method
(ICDSE), Cochin, India, 2016. for traffic time series imputation,” de Chinese Automation Congress
[10] S. Moritz, “Package ImputeTS,” cran.r-project.org, 2019. (CAC), 2017.
[11] S. Moritz, T. Bartz-Beielstein, “imputeTS: Time Series Missing Value [16] S. Moritz, A. Sardá, T. Bartz-Beielstein, M. Zaefferer, J. Stork,
Imputation in R,” The R Journal, vol. 9, nº 1, pp. 207-2018, 2017. “Comparison of different Methods for Univariate Time Series Imputation
in R,” arxiv.org, 2015.
[12] A. Chaudhry, W. Li, A. Basri, F. Patenaude, “On improving imputation
accuracy of LTE spectrum measurements data,” de Wireless [17] R. Little, “A test of missing completely at random for multivariate data
Telecommunications Symposium, Phoenix, AZ, USA, 2018. with missing values,” Journal of the American Statistical Association,
vol. 83, nº 404, pp. 1198-1202, 1988.
[13] S. Van Buuren, K. Groothuis-Oudshoorn, “mice: multivariate imputation
by chained equations in R,” Journal of Statistical Software, vol. 45, nº 3, [18] M. Khan, H. Hayat, I, Awan, “Hybrid case-base maintenance approach
2011. for modeling large scale case-based reasoning systems,” Human-centric
Computing and Information Sciences, vol. 9, nº 9, 2019.
382 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Worldwide Interoperability for Microwave Access added further modification to support higher reliability
network accepts the challenge of last mile wireless access for networks.
internet. IEEE 802.16 standard, commercially known as WiMAX
provide wireless broadband experience to the end subscribers Multi-hop wireless network started to get much focus in the
and challenges many wired solutions like Digital Subscriber Line telecommunication industry, due to rapid deployment and
(DSL) and cable internet. Wireless network has many inherent coverage enhancements. This also forces various product
issues like coverage holes; capacity optimization and mobility are portfolios to upgrade to this technology in order to get major
few of them. Adding relays to multi-hop WiMAX IEEE 802.16j share of the industry in near future. There are many networks
network present an effective solution to address them to some which provide multi-hop communication, but all of these have
extent but this amendment does not elaborate any algorithm different MAC and PHY layer design due to constraint like
regarding the relay selection and narrate no performance power, mobility and transmission range. Augmenting to this,
guarantees. In this work, we proposed linear model that fairly multi-hop wireless networks require complex routing
allocates wireless resources among subscribers in 802.16j algorithms. All this added to the limited use of the multi-hop
network. A relay selection algorithm is also presented to networks. Introduction of relays in multi-hop wireless network
optimally select nodes with higher signal-to-noise ratio as relay have improved the network capacity and extend coverage area
station for nodes with lower signal-to-noise ratio objectively and simplify deployment scenario. According to Pabst R. et al.
maximize overall network capacity. This scheme further extends
[6], relays provide communication to the nodes outside
network coverage area and improves network availability. We
transmission range of the base station (BS), support in
also did extensive performance evaluation of the proposed linear
model. Results show that optimal relays selection scheme do alleviating the range limitation for wireless broadband
provide a substantial increase of up to 66% in overall network networks.
capacity in the fixed WiMAX network. This improvement is There are many practical scenarios as shown in the Fig. 1,
substantial at places where network condition is not optimal. where relays do play important role. Relays can be placed in
Investigating the problem further leads to the conclusion that the different formations to enhance the system performance and
relay selection criterion is the key to achieve maximum network coverage. In fixed infrastructure the relays are placed in
capacity.
stationary areas by the service provider to provide normal
Keywords—WiMAX; multi-hop; wireless broadband; relay;
traffic by extenuating the loop holes and extending the
SNR coverage.
To provide relay based multi-hop support in WiMAX, a
I. INTRODUCTION new amendment to the standard is made known as IEEE
High speed internet access for the last mile has been a 802.16j and it defines two types of relays transparent and non-
challenge over the years. Apart from the wired technology, transparent relays. Transparent relays are used for
inception of advanced coding schemes and antenna technology improvement in capacity of overall network while the use of
made wireless broadband a competitive solution. IEEE 802.16 non-transparent relays extends the coverage area. The major
Working Group (2004) commercially known as WiMAX difference between the two relay modes is the way framing
initially formalized in 2001, however till 2004 standard based information is transmitted. In transparent mode frame header
on [1] targeted only fixed applications and the standard was information is not transmitted; while in non-transparent mode
referred to these as fixed WiMAX. In December 2005 new the frame contains the header information. The frame header
amendment IEEE 802.16e [2] was launched which added a encapsulates critical scheduling information which the nodes
new dimension in WiMAX, mobility. Since then IEEE used to determine when it can transmit or receive. In relay
802.16e-2005 forms the basis for the WiMAX solution for modes there are two types of scheduling modes: centralized
nomadic and mobile applications and is often referred to as and distributed. Base Station (BS) control scheduling for all the
mobile WiMAX. IEEE 802.16 [3] merges the fixed and mobile nodes in centralized mode while in distributed mode relay
PHY and MAC capabilities of the network. Another station (RS) can take some scheduling decision for the nodes
amendment IEEE 802.16 [4] modified the physical layer and attach to it. But there are challenges faced by the system
the MAC layer for inclusion of relays in the network known as implementing 802.16j like frequency reuse, resource
IEEE 802.16j. The current version of IEEE 802.16 [5] has allocation, relay selection, performance and scheduling. IEEE
383 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
802.16j standard has enhanced BS and RS capabilities to meet selection algorithm. The results and analysis of proposed
multihop networks communication challenges. The standard algorithm are detailed in Section 5. Section 6 concludes this
does not discuss much about relay selection criteria. This issue work.
has been left vendor specific. Similarly, no performance
parameters are given while using relays in the WiMAX II. RELATED WORK
network. In recent years, some research tried to predict the
performance of IEEE 802.16j network under different
realizations and addressed issues like performance of network
with inclusion of relays. Other issues include handling the loop
holes and actual coverage extension.
In recent work on the performance evaluation of the
WiMAX network, there are different attributes of wireless
network that are exploited to enhance the network throughput.
Like the work of Bonald T. et al. [10] is to determine the
maximum throughput gain achievable under max-min fairness
in which approximately equal performance is delivered to all
subscribers. Like the Genc V. et al. [16] present an analytical
modal that tries to enhance end-to-end throughput under max-
min fairness constraint. Further by Genc V. et al. [17] extends
Fig. 1. Relay Application.
the same model by incorporating the variable number of relays
and transmits power; results show that about 55% to 125% of
In this study we focus on the relay selection such that
the throughput gain can be achieved subject to spatial reuse.
throughput of overall network is optimized in IEEE 802.16j
But the MAXMIN algorithm has an issue that it may increase
network. In this context we propose linear model that fairly
the throughput of some subscribers but starved the rest of the
allocates the resources among subscribers in the range of BS.
nodes. Another way is to compare the SNR for each path as
In fixed and nomadic infrastructure relays with centralized
done by Shrestha D. M. et al. [11] purposes more advance
scheduling is the simplest and cost-effective solution to the
technique which use backward compatible signaling
multi-hop relay network. However distributed scheduling mode
mechanism and introduce a centralize path selection algorithm
is more affective for mobile applications in order to handle
based on ELT (estimated link throughput). ELT is based on
coverage extension and other mobility issues. Therefore, the
available bandwidth and data rate for maximum throughput
study focuses on centralized scheduling in fixed WiMAX
where data rate depends on SNR. ELT calculation for each
network. We develop linear model using linear programming
path and signaling make base station processing more complex
technique. We implemented the optimization model using
which compromises the overall performance of the network.
algebraic modeling problem language (AMPL) and use
simplex method for testing its results. We also develop a Similar work is purposed by Ann S. et al. [14] tries to find
simulation program in c programing language to test the the route that reduces the latency and maximizes the network
system model against IEEE 802.16-2004 based network throughput. It is a centralized algorithm that makes decision
without relays. Results show considerable throughput based on SNR, available bandwidth, hop count but this scheme
enhancement as compare to IEEE 802.16-2004. Further in this does uneven resource allocation and starves many subscribers.
research we consider two main aspects: (i) relay selection WiMAX provides two modes centralized and distributed
scheme in IEEE 802.16j network, (ii) implications of relays on scheduling. Some research uses centralized mode to achieve
the overall network performance. Followings are the the performance as in Li D. et al. [18] focus on the maximum
contributions of this work: throughput in two hop fixed WiMAX network by using
centralized scheduling scheme. Relay selection algorithm in
We first propose a linear model to evaluate the such proposal is based on quality-of-service QoS, parameters
performance of WiMAX network. including SNR and latency. But it did not warrant throughput
We also develop relays selection algorithm for WiMAX maximization that can be achieved using this algorithm. One of
multi-hop network and perform optimization of the techniques is to measure the resource cost for the path, in
proposed linear model. order to select the most optimum path for the channel like the
work in [20] proposes a path selection model based on radio
We then perform comprehensive performance analysis resource cost (RRC) and the minimum the cost the best is the
of capacity gains after inclusion of relays in the path. The result shows the throughput obtain using RRC
WiMAX multi-hop network. scheme is much better as compare to the rest of scenarios. An
Results show that proposed relay-based model out interesting study use the adaptive technique according to the
performs traditional WiMAX network in terms of channel conditions as in Chang J. Y. et al. in [22] develop a
capacity utilization. deployment algorithm that work based on traffic and uniform
clustering. This study takes both RS and BS placement to get
The rest of the paper is organized as follows. In Section 2, the best throughput and coverage. The algorithm provides the
related work has been elaborated. Section 3 presents the system vendors an adaptive deployment of BS and RS, further
model and problem formulation. Section 4 discusses the relay considering the environment.
384 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Liu I. H. et al. [23] performs performance analysis and compared to the RSPS and RPCC schemes. The average
purpose scheduling algorithm that suggests that if the boundary throughput and coverage ratio show profound improvement.
for the zones are kept adaptive then the overall performance of
the network can be enhanced notably. In support of this another Work done by Arthi M. et al. [25] mainly focuses on the
study by Rajya Lakshmi et al. [28] improves the performance IEEE 802.16j network RS placement. The author emphasis the
of MMR WiMAX networks and maintains the QoS flow need of proper placement of RS in the network because
requirements by using adaptive Zone size based on channel improper placement may introduce multitude of issues like
quality of each node. But if the number of subscribers increases SNR in case of densely use of RS nodes, transmission delay in
with more variable channel conditions the purposed method case of long spread RS nodes, Coverage holes, path selection
will be expensive to use and may affect the overall in case of more than one option available for any node finally
performance. Even the power mode can be used for the the link overloading such as many nodes requesting
performance improvement in the overall network as shown by communication from one RS node. The paper selects the
the research by Paul A. et al. [24]; exploit sleep mode candidate positions for the RS using throughput-oriented
parameters of IEEE 802.16j to control or minimize the energy method. Then develop an optimized model for the deployment
needs of the mobile node. They propose a scheme called of the RS in the network in order to minimize the overall
energy saving centric uplink scheduling ESCS. This scheme budget. The model simulation result shows quite improvement
does bandwidth allocation and sleep cycle decision algorithm. in the capacity of the network. Especially when the BS an RS
The results show that proposed ESCS provide more sleep time distance increases till a point where one can get the maximum
to the nodes, hence making them more energy efficient. throughput. These points become the deployment place for the
Interference is the major constraint for the wide spread shared RS. Relay selection is also important as the relay placement.
resources in WiMAX network among the devices and become Relay selection plays very important role in enhancing the
challenge for the researchers. Therefore, resource allocation overall network performance as shown by the study Sreng V. et
and sharing become an issue. In Mahb. et al develop a greedy al. [7] that propose relay selection strategy based on physical
centralized algorithm handle this issue. Another concept distance and path-loss and conclude with the results that path-
introduce relays in the network comes into the lime light. loss selection scheme consistently shows superior performance
Initially relays were used to extend the network coverage. But in comparison to those based on distance only. However, the
IEEE 802.16j made relays more useful in the context of study only discusses the coverage extension; impact on
improving the performance and QoS of the network. network throughput is not considered. Researchers devise
The main question is do the relays in network can improve different parameters for the selection of relays, some purpose
the overall network performance or otherwise? For this methods such as selection based on distance from BS, simplest
research get focus on this aspect like study by Deb S. et al. [13] ways in this issue. Some studies find out that SNR (signal-to-
demonstrated the enhancement in throughput and range noise ratio) should be considered as a main factor for the relay
extension using relays at MAC layer. The results show an selection, therefore Hu H. et al. [8] concludes that station with
improvement in the median throughput of about 25%. But their maximal SNR in candidate relay stations should be chosen as
analysis addresses only downlink scheduling however uplink relay. Some selection is based on the power criteria aspect
results can be obtained similarly. Also, the relay selection for considered by the researchers for the relay selection, the
the respective subscriber remains unanswered in this work. A algorithm proposed by Hui T. et al. [9] choose the minimal
similar study by Genc V. et al. [15] shows that introduction of total transmission power, set P1 and P2 for the power of the
relays only improves throughput of approximately half of the relay link and access link respectively, this find the complete
coverage area of base station (BS). path and relay will be selected automatically. Some study
selects the different modes of relays for the performance
Next question is, will the relay placement and selection improvement of the network. Such as Zhu V. et al. [12] did
play any role in the overall network performance. First, if we comprehensive performance evaluations of relays in WiMAX
consider the relay placement lot of work shows the importance networks and finds out that the non-transparent relay station in
of relay placement in the context of the network overall IEEE 802.16j network do perform better with distributed
performance like work by Chang C. Y. et al. [26] purposes a scheduling as compare to the transparent relays. But non-
relay placement mechanism (RPM) that enhances the overall transparent relays are expensive and complex, and therefore
network capacity. Simulation results shows improvement in most research studies are done for transparent relays which are
throughput and delays while maintaining a satisfactory level more economical and easier to deploy. Another relay selection
QoS. Also, performance evaluation in MMR networks is criterion is the mode of working principle of the relays like
carried out by Ge Y. et al. [19] and they show that optimal amplify-forward (AF) and decode and forward (DF) relays as a
relay selection can enhance the end-to-end throughput up to Study by Swain Manoj et al. [29] propose relay selection
approximately 49%. Further Chang C. Y. et al. in [21] propose scheme based on both AF and DF types of relays. The results
a relay deployment algorithm with an objective to minimize the show that the harmonic mean is better than min max scheme as
hardware cost. Result shows improvement in the time slots far as SNR requirements for selection of relays is concern but
allocation when relays are optimally placed but still for bandwidth performance min max supersedes harmonic
compromise regarding hardware cost. In Chang J. Y. et al. in mean. However, the results are simulated keeping fixed
[27] proposed relay placement in order to improve network number of subscribers. In [30], we tried to maximize network
performance by minimizing the budget using rotational capacity by optimizing relay selection.
clustering algorithm. The results of the purpose scheme are
385 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
mij Data rate per slot against a specific MCS for the
subscriber between node i and j
J Base station node
386 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
station and the relay node, a relay node can be any node in the
network except the requesting subscriber. F. Resource Constraint
In the world of wireless data communication resources are
We used simplex method for the maximization function, the most precious and scarce thing. In proposed modal the
initially the function allocate slot for the subscriber over the resource we consider is the physical slots. Since q represent the
link between the base station and the subscriber, then further total number of slots in a downlink frame. Therefore, it is
allocates the slots to the relay nodes for the given subscriber checked that the total number of slots allocated to all the
over the link between base station and the relays in the same subscribers and relays should be equal or less than the total
frame. The function selects the relay nodes for the given number of slots available i.e. q. In Eq. (4), Ts e js represent the
subscribers if the number of slots allocated to the subscriber in
the direct link to the base station is more than the number of number of slots allocated to the subscriber over the direct link
slots via relay nodes. The equation Ts e jk .m jk is multiplied by two between the base station and the subscriber. Ts e jk shows the
s
s s
and subtracted by number slots allocated to the relay nodes over the link between
Ts e k s .m k s to show two hop communication.
s s the base station and the relay node. Ts ek s here the Ts represents
s
Since Ts e k s .m k s shows the data rate allocated to the relays which
s s the number of slots allocated to the subscriber attach to the
need to be subtracted to get the actual data rate of the relay ks. Summation of all the allocated slots to all the nodes
subscribers attach to that relay. We assume centralized over all the links should be less or equal to q.
scheduling; therefore, base station allocates slots both for the
relay and subscriber. Proposed objective is to maximize the T e T e
s js s jks T e s ks s q
overall data rate of the network under given constraints. These s s ks s ks
(4)
constraints are explained one by one as below:
j J , ks K s , s S
T e s js .m js 2T e s jks .m jks T e s ks s .mks s
s s ks s ks
(1) G. Share Constraint
j J , ks K s , s S In order to make our model fair, following equations are
constructed. Here in Eq. (5), u is the total number of
D. Demand Constraint subscribers in a network. (q / u).m js gives average data rate for
First constraint is obvious that the actual data rate allocated each subscriber. (q / u).m js d s this equation checks that if the
to each subscriber must be equal or less than its actual demand.
Since the demand of the subscriber does vary therefore to keep demand of subscriber is more than the average data rate
the system safe from abnormal allocations this constraint plays available for it or vice versa. In case of more demand Eq. (5)
will be executed. Eq. (6) ensures that each subscriber should
a vital role. In Eq. (2), Ts e js .m js shows data rate of the
get at least average data rate or more if possible. In the next
subscriber for direct link between base station and the section we compare our model with fixed WiMAX. The
subscriber. Similarly, Ts e jks .m jks shows the data rate for the simulation results show substantial improvement in the overall
subscriber over the indirect link via relay. If the subscriber network throughput. These results with given scenarios are
attaches to the base station directly then the relay link will have further elaborated below:
Ts with zero value, which further makes the data rate (q / u).m js d s 0
insignificant. Therefore, numerically we will be getting the data (5)
rate of only one link, which in our scenario is the direct one.
j J,sS
Ts e js .m js e
ks
jks .m jks ds
Ts e js .m js T e s jk s .m jk s ds 0
(2) ks
(6)
j J , ks K s , s S
j J , ks K s , s S
387 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
assign to all the nodes must not surpass the given maximum
number of slots. Finally the share constraint insert fairness V. RESULTS AND DISCUSSION
attribute in the algorithm Tas .m js Trs .m jks d s 0 this For the simulation purpose we employed tree-based
equation forces the algorithm to assign some average resource topology with BS at the root and centralized scheduling is
to the node. Next section will discuss the results we obtain assumed. Also, nodes only two hops away from the BS are
from this algorithm. taken under consideration. Since more than two hops have
capacity issues depend upon subscriber density. We create
TABLE. II. ALGORITHM PARAMETERS different downlink scenarios to test the effectiveness of our
proposed model. The scenario we assume is such that on
Parameter Description OFDMA PHY using frequency 3.5GHz, 20MHz channel
bandwidth, size is 2048 FFT, contain 1440 data subcarrier and
J Base Station
1/8 cyclic prefix. In addition to this other parameter like noise
S Set of all nodes figure and thermal noise at transmitter are adjusted at 13db and
ks Set of all nodes except s -174dbm respectively. The frame size is assumed to be 20ms.
The DL sub frame is about uses 50 of the total frame sizes.
mij Set of all MCS between Base station and nodes S
Major parameter that we take under consideration for
m jks Set of all MCS between Base station and nodes Ks analysis purpose is SNR [signal to noise ratio]. SNR value
determines the quality of a link ranges from 1-26, if its value is
mks s Set of all MCS between nodes S and Ks high then link is good and can transmit data at higher rate
because of better modulation and coding scheme and vice
ds Demand of node s versa. Our focus in this simulation is to cover all the possible
Tas number of nodes assign over the link BS and s node scenarios that come across in networks. Table III shows the
quantitative values of SNR against which the respective
Trs number of nodes assign over the link BS and Ks node modulation and coding scheme is selected. In this we see that
as the values of SNR increases, the better MCS are mapped
Tak number of nodes assign over the link Ks and s node
against them.
388 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. III. SIGNAL TO NOISE RATIO throughput of the network by giving alternate routes to the
subscribers with low or poor SNR values.
Modulation Coding Scheme Receiver (SNR)
389 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 7. 100% Subscribers with Excellent SNR. Fig. 8. 100% Subscribers with Poor SNR.
390 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[2] IEEE LAN/MAN Standards Committee. IEEE Standard for local and [16] Genc, Vasken, Sean Murphy, and John Murphy. "An interference-aware
metropolitan area networks Part 16: Air interface for fixed and mobile analytical model for performance analysis of transparent mode 802.16 j
broadband wireless access systems amendment 2: Physical and medium systems." In 2008 IEEE Globecom Workshops, pp. 1-6. IEEE, 2008.
access control layers for combined fixed and mobile operation in [17] Genc, Vasken, Seán Murphy, and John Murphy. "Analysis of
licensed bands and corrigendum 1. IEEE Std 802.16 e-2005. 2006. transparent mode IEEE 802.16 j system performance with varying
[3] IEEE 802.16 Working Group. Ieee standard for local and metropolitan numbers of relays and associated transmit power." In 2009 IEEE
area networks-part 16: Air interface for fixed broad-band wireless access Wireless Communications and Networking Conference, pp. 1-6. IEEE,
systems. IEEE Std. 802.16-2004. 2004. 2009.
[4] IEEE 802.16 Working Group. IEEE standard for local and metropolitan [18] Li, Dandan, and Hao Jin. "Relay selection in two-hop IEEE 802.16
area networks, part 16: Air interface for broadband wireless access Mobile Multi-hop Relay networks." In 2009 First International
systems, amendment 1: Multi-hop relay specification. IEEE Standard Workshop on Education Technology and Computer Science, vol. 2, pp.
802.16 j-2009. 2009. 1007-1011. IEEE, 2009.
[5] Pareit D, Lannoo B, Moerman I, Demeester P. The history of WiMAX: [19] Ge, Yu, Su Wen, and Yew-Hock Ang. "Analysis of optimal relay
A complete survey of the evolution in certification and standardization selection in IEEE 802.16 multihop relay networks." In 2009 IEEE
for IEEE 802.16 and WiMAX. IEEE Communications Surveys & Wireless Communications and Networking Conference, pp. 1-6. IEEE,
Tutorials. 2011 Oct 13;14(4):1183-211. 2009.
[6] Pabst, Ralf, Bernhard H. Walke, Daniel C. Schultz, Patrick Herhold, [20] Mach, Pavel, Robert Bestak, and Zdenek Becvar. "Optimization of
Halim Yanikomeroglu, Sayandev Mukherjee, Harish Viswanathan et al. association procedure in WiMAX networks with relay stations."
"Relay-based deployment concepts for wireless and mobile broadband Telecommunication Systems 52, no. 3, 2013.
radio." IEEE Communications Magazine 42, no. 9 (2004): 80-89. [21] Chang, Chih-Yung, and Ming-Hsien Li. "A placement mechanism for
[7] Sreng, Van, Halim Yanikomeroglu, and David D. Falconer. "Relayer relay stations in 802.16 j WiMAX networks." Wireless networks 20, no.
selection strategies in cellular networks with peer-to-peer relaying." In 2, 2014.
2003 IEEE 58th Vehicular Technology Conference. VTC 2003-Fall [22] Chang, Jau-Yang, and Ya-Sian Lin. "A clustering deployment scheme
(IEEE Cat. No. 03CH37484), vol. 3, pp. 1949-1953. IEEE, 2003. for base stations and relay stations in multi-hop relay networks."
[8] Hu, Huining, Halim Yanikomeroglu, David D. Falconer, and Shalini Computers & Electrical Engineering 40, no. 2, 2014.
Periyalwar. "Range extension without capacity penalty in cellular [23] Liu, I-Hsien, Chuan-Gang Liu, Chien-Tung Lu, Yi-Tsen Kuo, and Jung-
networks with digital fixed relays." In IEEE Global Telecommunications Shian Li. "A multi-hop resource scheduling algorithm for IEEE 802.16 j
Conference, 2004. GLOBECOM'04., vol. 5, pp. 3053-3057. IEEE, 2004. relay networks." Computer Networks 67, 2014.
[9] Hui, Tian, Gu Xuelin, and Zhang Ping. "The impact of relaying [24] Paul, A., Anagha, P., and Umaparvathi, M. "Energy Efficient
strategies on the performance in cellular system." In IEEE International Scheduling For Wimax Network", International Journal of Software &
Symposium on Communications and Information Technology, 2005. Hardware Research in Engineering, 3(3), 2015.
ISCIT 2005., vol. 2, pp. 1404-1407. IEEE, 2005.
[25] Arthi, M., Jimy Jose Joy, P. Arulmozhivarman, and K. Vinoth Babu.
[10] Bonald, Thomas, Laurent Massoulié, Alexandre Proutiere, and Jorma "An efficient relay station deployment scheme based on the coverage
Virtamo. "A queueing analysis of max-min fairness, proportional and budget constraints in multi-hop relay networks." In 2015
fairness and balanced fairness." Queueing systems 53, no. 1-2 (2006): International Conference on Communications and Signal Processing
65-84. (ICCSP), pp. 0124-0128. IEEE, 2015.
[11] Shrestha, Deepesh Man, Sung-Hee Lee, Sung-Chan Kim, and Young- [26] Chang, Chih-Yung, Chao-Tsun Chang, Tzu-Chia Wang, and Ming-
Bae Ko. "New approaches for relay selection in IEEE 802.16 mobile Hsien Li. "Throughput-enhanced relay placement mechanism in
multi-hop relay networks." In European Conference on Parallel WiMAX 802.16 j multihop relay networks." IEEE systems journal 9, no.
Processing, pp. 950-959. Springer, Berlin, Heidelberg, 2007. 3, 2014.
[12] Zhu VM, Viorel VD. Multihop relay extension for WiMAX networks— [27] Chang, Jau-Yang, and Yun-Wei Chen. "A relay station deployment
overview and benefits of IEEE 802.16 j standard. Fujitsu Sci. Tech. J. scheme with a rotational clustering algorithm for multi-hop relay
2008 Jul;44(3):292-302. networks." In 2016 International Conference on System Science and
[13] Deb, Supratim, Vivek Mhatre, and Venkatesh Ramaiyan. "WiMAX Engineering (ICSSE), pp. 1-4. IEEE, 2016.
relay networks: opportunistic scheduling to exploit multiuser diversity [28] Lakshmi, L. Rajya. "Adaptive Zone Size Selection Method for IEEE
and frequency selectivity." In Proceedings of the 14th ACM 802.16 j Mobile Multihop Relay Networks." Wireless Personal
international conference on Mobile computing and networking, pp. 163- Communications 97, no. 4, 2017.
174. ACM, 2008.
[29] Swain, Chaudhuri Manoj Kumar, and Susmita Das. "Study and Impact
[14] Ann, Sojeong, Kyung Geun Lee, and Hyung Seok Kim. "A path of Relay Selection Schemes on Performance of an IEEE 802.16 j Mobile
selection method in IEEE 802.16 j mobile multi-hop relay networks." In Multihop Relay (MMR) WiMAX Network." In Progress in Intelligent
2008 Second International Conference on Sensor Technologies and Computing Techniques: Theory, Practice, and Applications, pp. 491-
Applications (sensorcomm 2008), pp. 808-812. IEEE, 2008. 499. Springer, Singapore, 2018.
[15] Genc, Vasken, Seán Murphy, and John Murphy. "Performance analysis [30] N. Mazhar, “802.16j Network Performance Analysis and Relay
of transparent relays in 802.16 j MMR networks." In 2008 6th Selection” MS. dissertation, Dept. Computing, SEECS, NUST Univ.,
International Symposium on Modeling and Optimization in Mobile, Ad Islamabad, Pakistan, 2011.
Hoc, and Wireless Networks and Workshops, pp. 273-281. IEEE, 2008.
391 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Since health and the quality of life are caused not persists, the proportion of the elderly will reach 20.8% in 2026
by a single factor but by the interaction of multiple factors, it is and South Koreans will enter a post-aged society.
necessary to develop a model that can predict the quality of life
using multiple risk factors rather than to identify individual risk Since nuclearized families, decreased socioeconomic
factors. This study aimed to develop a model predicting the capacities, and chronic degenerative diseases such as dementia
quality of life based on C-SVM using big data and provide have increased, the quality of life in old age is lower than that
baseline data for a successful old age. This study selected 2,420 of the young and middle-aged people [2]. The elderly are
elderly (1,110 men, 1,310 women) who were 65 years or older and particularly vulnerable to diseases. Recent studies [3, 4, 5]
completed the Seoul Statistics Survey. The quality of life reported that communication issues (e.g., dementia and
satisfaction, a binary outcome variable (satisfied or dissatisfied), aphasia) and swallowing problems as well as physical
was evaluated based on a self-report questionnaire. This study problems (e.g., degenerative joint diseases) adversely affected
performed a Gauss function among the SVM algorithms. To the quality of life in old age. The physical aging and chronic
verify the predictive power of the developed model, this study diseases that people experience in old age shrink the elderly
compared the Gauss function with the linear algorithm, psychologically and deprive them of positive emotions [6].
polynomial algorithm, and sigmoid algorithm. Additionally, C- Additionally, psychological aging decreases mental and
SVM and Nu-SVM were applied to four kernel algorithm types neurological functions such as sensation and intelligence and it
to create eight types, and prediction accuracies of the eight SVM
also leads to emotional changes such as anxiety and depression
types were estimated and compared. Among 2,420 subjects, 483
[6]. Furthermore, the elderly often suffer from various
elderly (19.9%) were not satisfied with their current lives. The
final prediction accuracy of this SVM using 625 support vectors difficulties such as aggravated health, disabilities, higher
was 92.63%. The results showed that the difference between C- psychological alienation and loneliness, decreased economic
SVM and Nu-SVM was negligible in the models for predicting ability, reduced the social role, and declined informal network
the satisfaction of life in old age while the Gaussian kernel had due to diverse factors including physical aging, chronic
the highest accuracy and the sigmoid kernel had the lowest diseases, and loss of spouse [7].
accuracy. Based on the prediction model of this study, it is The satisfaction of life is a subjective evaluation regarding
required to manage local communities systematically to enhance
the satisfaction and perception of one’s current life [8].
the quality of life in old age.
Moreover, this concept is widely used in gerontology and
Keywords—C-SVM; communication frequency; life geriatrics studies [8]. An individual’s quality of life indicates
satisfaction; social network; quality of life the well-being state containing the concepts of satisfaction,
happiness, and positive emotions and includes physical factors,
I. INTRODUCTION mental factors, social factors, and personal achievements [9].
Globally, the proportion of the elderly population is rapidly In other words, the quality of life is an index that reflects not
increasing due to economic growth, the advancement of only physical factors but also mental factors such as health,
medical technologies, and improved living standards. The occupational factors, social interaction, happiness, and
aging speed of South Korea, in particular, is much faster than satisfaction. The satisfaction of life is an important issue in
other countries in Europe, North America, Oceania, and Africa. gerontology because the satisfaction of life is not obtained by
As of 2018, the number of the elderly aged 65 or over was 7.38 the general conditions of the elderly but formed by the
million (14.3% of the total population), indicating that one out interaction between the physical environment and the social
of 7 people in the population is a senior citizen [1]. If this trend environment that are directly experienced by them.
*Corresponding Authors.
392 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Previous studies [10, 11, 12] have identified a variety of in the welfare level by sector and year, and secure basic data
factors affecting the quality of life in old age such as for establishing the welfare policies of Seoul. The purposes of
sociodemographic factors (e.g., gender, age, income level, and the Seoul Welfare Panel Survey were to (1) measure the
educational level), marital status, and chronic diseases. Wada welfare level of the resident population of Seoul and track the
et al. [13] reported that the subjective satisfaction of life changes in welfare level by year; (2) estimate the demand for
correlated with physical functions and health. Recent studies welfare services by identifying the size and status of the
showed that social networks and the frequency of contacting socially vulnerable class; (3) measure the effects of the
people around them also influenced the quality of life in old program by evaluating the accessibility and satisfaction of
age [14, 15]. The elderly had higher satisfaction in life when citizens for the existing welfare service programs; (4) present
they contacted family, relatives, friends, and neighbors more baseline data for providing comprehensive welfare program
frequently and when they had various activities such as using through the above survey and measurement; and (5) pave the
senior citizen centers and community welfare centers [16]. way to revitalize social science studies based on quantitative
data by providing statistical data to social welfare researchers.
Since health and the quality of life are caused not by a The Seoul Statistics Survey contains items related to the
single factor but by the interaction of multiple factors, it is quality of life such as income, consumption, savings, debt,
necessary to develop a model that can predict the quality of life assets, living conditions, health, housing, elderly support, child
using multiple risk factors rather than to identify individual risk education, disability and rehabilitation, welfare services,
factors [17]. Nevertheless, previous studies [18, 19] evaluating cultural lives, and social participation. This study used
the risk factors affecting the quality of life in old age just aimed computer assisted personal interviewing for surveying that
to explore individual risk factors using the generalized linear trained investigators visited the homes of subjects and
model. These studies [18, 19] mostly used logistic regression conducted face-to-face interviews using laptops. This study
models to identify predictors. However, since regression selected 2,420 adults (1,110 men, 1,310 women) who were 65
models are used to predict the strength of the relationship years or older and completed the Seoul Statistics Survey as the
between independent variables and dependent variables, this final subjects.
method is not appropriate to discover new predictors.
B. Measurements of Variables
Recently, supervised learning algorithms such as the
support vector machine (SVM) have been widely used in social The definitions of the variables measured in this study are
science as a method of identifying complex factors associated presented in Table I. The quality of life satisfaction, a binary
with diseases and health problems [20]. Numerous studies [21, outcome variable (satisfied or dissatisfied), was evaluated
22] have reported that the SVM has higher prediction power based on a question, “Are you satisfied with your life?”.
(accuracy) in classifying or predicting binary data than decision Explanatory variables included gender, age (i.e., 65-74 years
tree based machine learning or logistic models. Particularly, old and 75 years or older), the highest level of education (i.e.,
[23] showed that C-SVM, a transformation algorithm of the below elementary school, middle school, high school, and
SVM, performs better because it makes the classification equal to or higher than collage graduation), monthly mean total
margin to classify two categories a serpentine nonlinear shape. household income (i.e., <2 million KRW, 2 million KRW≤
This hyperparameter technique is drawing attention as a way to and <4 million KRW, and 4 million KRW≤), marital status
increase prediction power. (i.e., living with a spouse, married but not living with a spouse,
and not married), current employment status (i.e., employment
To date, it has not been tried to develop a prediction model
and unemployment), drinking frequency (i.e., less than once a
reflecting health habits, subjective health, communication
week and more than once), smoking (i.e., smoking and non-
frequency, and social network, in addition to demographic
smoking), walking per week (i.e., equal to or more than two
factors, based on the supervised learning algorithm. This study
days and less than one day), subjective health status (i.e., good,
aimed to develop a model predicting the quality of life based
normal, and bad), depression (i.e., yes and no),, disease or
on C-SVM using big data and provide baseline data for a
accident experience in the past two weeks (i.e., yes and no), the
successful old age. Construction of our study is as follows.
frequency of meeting a neighbor (i.e., less than once a month
Section II explains database and analyzed variables and
and more than twice a month), and frequency meeting a
Section III defines C-SVM and explains the procedure of
relative (i.e., less than once a month and more than twice a
model development. Lastly, Section IV presents conclusion for
month). Depression was measured using the Short Form
future studies.
Geriatric Depression Scale (SGDS) [24]. SGDS is easy to test
II. METHODS AND MATERIALS the depression of the elderly, does not take much time, and is
highly valid [24]. For each question, 'Yes' was given 1 point,
A. Target Subjects and 'No' was given 0 points. The total depression score ranged
The data source of this study was the 2015 Seoul Statistics from 0 to 15. The threshold for depression was 8, and a higher
Survey. The Seoul Statistics Survey was conducted with the score means more severe depression. The reliability of SGDS
permission of the Statistics Korea in order to measure the in this study was evaluated by Cronbach's α (.881).
welfare level of Seoul's resident population, track the changes
393 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Frequency meeting a More than twice a month, Fig. 1. The Concept of Kernel Functions [26].
relative Less than once a month
Disease or accident
Health status experience in the past No, Yes
two weeks
Short Form Geriatric
Depression No, Yes
Depression Scale(SGDS)
Subjective Life
Life satisfaction Satisfied, not satisfied
Satisfaction
394 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
395 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. III. VALUES OF FUNCTION WEIGHTS TABLE. IV. THE ACCURACY OF PREDICTING THE SATISFACTION OF LIFE
ACCORDING TO THE CLASSIFICATION ALGORITHM OF THE SVM, %
Age
65-74 .011 Type of Type of Kernel
SVM Linear Polynomial, Gaussian Sigmoid
75+ -.032
Gender C-SVM 91.25 90.58 92.63 89.88
Male -.019 Nu-SVM 91.14 90.77 92.10 88.85
Female .005
Education V. DISCUSSION
Elementary school graduation and
-.030 This study developed an SVM-based model for predicting
below the satisfaction of life in old age using the data of the Seoul
Middle school graduation -.028 Statistics Survey. It was found that 20% of the surveyed elderly
High school graduation .003 were dissatisfied with their current lives. “The Satisfaction of
College graduation and above .007 Life of Elderly Population Groups [27]” published by Statistics
Economic activity
Korea in 2018 showed that the satisfaction of life (≥50 years
old) of South Korean was 5.4 out of 11 points, which was 1
Employed .025
point lower than the average of OECD countries (6.4 points),
Not-employed -.033 even though it compared with cannot be directly the results of
Household Income this study. It was ranked as 28thamong35OEC
Below 2 million KRW -.007 Dmembercountries:2pointslowerthanDenmark (7.6 points; the
Between 2 and 4 million KRW -.007 top score), and even lower than Japan (5.8 points) [27]. If
4 million KRW or above .005 South Korean enters a post-aged society with continuing this
trend, it is more likely to decrease the quality of life in old age.
Spouse
Therefore, it is necessary to prepare polices at a society level
Cohabitation .017 and take active measures to improve the quality of life in old
Separation -.027 age.
Bereavement .019
The results of this study revealed that health status (e.g.,
Drinking frequency
subjective health status, depression, and the experience of a
Once a week or less .030 disease or accident in the past two weeks), health habits
Once a week or more .028 (number of walking per week), and social network/
Smoking communication frequency (the frequency of meeting a
Current smoker .007 neighbor and the frequency of meeting a relative) were main
Non-smoker .009
factors for predicting the quality of life in old age in addition to
demographic factors (e.g., age, gender, education, economic
Subjective health status
activity, and marital status). Numerous previous studies
Good .007 evaluating the quality of life in old age reported that health
Average .004 status [30] and marital status [31] were main factors affecting
Poor -.011 the quality of life and these results, in addition to
Walking per week sociodemographic characteristics such as economic level [28]
One day or more .020
and age [29], agree with the results of this study. Previous
studies on the satisfaction of life related to the economic
Less than one day -.034
stability of the elderly [32, 33] showed that those with low
Depression incomes generally were recipients of national basic livelihood
Yes -.037 guarantees and they were not satisfied with their living
No .018 conditions when they did not have a spouse. As of 2018, the
Disease or accident experience in the poverty rate of the elderly (≥65 years) in South Korea was
past two weeks 48.8%, which is four-folds of the average (12.1%) of OECD
Yes -.023 countries [27]. Therefore, economic support is needed to
No .015 improve the quality of life in old age.
Frequency of meeting a neighbor Previous studies [32, 33, 34, 35] indicated that the better
Less than once a month -.038 health status of the elderly increased the level of their life
Twice or more per month .015 satisfaction while functional impairment (e.g., communication
Frequency of meeting a relative
problem) that adversely influenced the interaction with others
negatively affected the satisfaction of life. In particular, it was
Less than once a month -.031
found that health status had a greater effect on declining life
Twice or more per month .018 satisfaction in female elderly than male elderly and older
Number of support vectors: 625 elderly than younger elderly [33]. Additionally, many studies
showed that marital status was a major factor affecting the
396 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
quality of life [34]. The elderly living with spouses had a health status and well-being. The Annals of Family Medicine, vol.2,
higher level of life satisfaction than those who were unmarried no.6, pp.555-562, 2004.
or those who lose their spouses [35]. Particularly, marital status [7] J. B. Unger, G. McAvay, M. L. Bruce, L. Berkman, and T. Seeman,
Variation in the impact of social network characteristics on physical
was a very important factor in determining life satisfaction for functioning in elderly persons: MacArthur Studies of Successful Aging.
older elderly than younger elderly. The Journals of Gerontology Series B: Psychological Sciences and
Social Sciences, vol. 54, no. 5, pp. S245-S251, 1999.
An interesting finding of this study was that social
[8] E. Diener, R. E. Lucas, and S. Oishi, Subjective well-being: The science
networks and communication frequency were major predictors of happiness and life satisfaction. Handbook of positive psychology, vol.
for the quality of life in old age. Kim et al. [36] also reported 2, pp. 63-73, 2002.
that the satisfaction of life increased with more frequent [9] U. Schimmack, P. Radhakrishnan, S. Oishi, V. Dzokoto, and S. Ahadi,
contacts and higher quality contacts with family members, Culture, personality, and subjective well-being: Integrating process
friends, and neighbors (so-called better social networks). It models of life satisfaction. Journal of personality and social psychology,
could be because the elderly would require more social support vol. 82, no. 4, 582-593, 2002.
than younger people in order to relieve loneliness and physical [10] J. Butler, and J. Ciarrochi, Psychological acceptance and quality of life
in the elderly. Quality of life Research, vol. 16, no.4, pp. 607-615, 2007.
unwellness [36]. It is particularly well known that emotional
support from the family has an important effect on resolving [11] S. Y. Park, A study on depression, ADL, IADL, and QOL among
community-dwelling, low income elderly. Journal of Korean Public
the loneliness of the elderly [16]. It has been reported that the Health Nursing, vol. 23, no. 1, pp. 78-90, 2009.
elderly who receive more support from the family have better [12] A. Öztürk, T. T. Şimşek, E. T. Yümin, M. Sertel, and M. Yümin, The
emotional health, better life satisfaction, and lower loneliness relationship between physical, functional capacity and quality of life
[36]. Therefore, in order to improve the quality of life in old (QoL) among elderly people with a chronic disease. Archives of
age, it is necessary to establish a system that can increase the Gerontology and Geriatrics, vol. 53, no. 3, pp. 278-283, 2011.
frequency of regular contacts (communication) with the [13] T. Wada, M. Ishine, T. Sakagami, K. Okumiya, M. Fujisawa, S.
members of local community groups such as volunteers as well Murakami, K. Otsuka, S. Yano, T. Kita, and K. Matsubayashi,
Depression in Japanese community-dwelling elderly—prevalence and
as relatives, friends, and neighbors. association with ADL and QOL. Archives of Gerontology and
Geriatrics, vol. 39, no.1, pp, 15-23, 2004.
Another major finding of this study was that the prediction
accuracy of C-SVM’s Gaussian kernel was higher than that of [14] H. Byeon, Developing a model to predict the social activity participation
of the senior citizens living in South Korea by combining artificial
linear kernel, polynomial kernel, and sigmoid kernel neural network and quest algorithm. International Journal of
algorithms. The performance of nonlinear SVM depends on the Engineering & Technology, vol. 8, no. 1.4, pp. 214-221, 2019.
kernel functions applied to the algorithms and parameters [15] A. P. Lane, C. H. Wong, Š. Močnik, S. Song, and B. Yuen, Association
composing them. The Gaussian kernel is an algorithm that of Neighborhood Social Capital With Quality of Life Among Older
maps the data to a characteristic space of infinite dimension. People in Singapore. Journal of Aging and Health, vol. e-pub:
Author in [21] also proved that it is an algorithm with high doi.org/10.1177/0898264319857990, 2019.
prediction accuracy. Therefore, it is believed that using a [16] O. M. R. Gouveia, A. D. Matos, and M. J. Schouten, Social networks
and quality of life of elderly persons: a review and critical analysis of
Gaussian kernel based C-SVM algorithm will be more literature. Revista Brasileira de Geriatria e Gerontologia, vol. 19, no. 6,
effective for subsequent studies to develop models for pp. 1030-1040, 2016.
predicting binary variables than using a sigmoid algorithm. [17] H. Byeon, and S. Kim, Development of risk prediction model for stroke
among Korean older adults using quest algorithm: a community-based
The results of this study would provide an important basis cross-sectional study. International Journal of Applied Engineering
that must be considered for developing health policies for Research, vol. 10, no. 79, pp. 93-96, 2015.
successful aging. Based on the prediction model of this study, [18] S. G. Kumar, A. Majumdar, and G. P, Quality of life (QOL) and its
it is needed to manage local communities systematically to associated factors using WHOQOL-BREF among elderly in urban
enhance the quality of life in old age. Puducherry, India. Journal of clinical and diagnostic research: JCDR,
vol. 8, no. 1, pp. 54-57, 2014.
REFERENCES [19] J. Butler, and J. Ciarrochi, Psychological acceptance and quality of life
[1] Statistics Korea, statistics of elderly persons, Statistics Korea, Daejeon, in the elderly. Quality of life Research, vol. 16, no. 4, pp. 607-615, 2007.
2018. [20] S. Khan, and T. Yairi, A review on the application of deep learning in
[2] A. Bowling, Ageing well: Quality of life in old age, McGraw-Hill system health management. Mechanical Systems and Signal Processing,
Education, London, 2005. vol. 107, pp. 241-265, 2018.
[3] P. H. Chen, J. S. Golub, E. R. Hapner, and M. M. Johns, Prevalence of [21] H. Byeon, Model development for predicting the occurrence of benign
perceived dysphagia and quality-of-life impairment in a geriatric laryngeal lesions using support vector machine: focusing on South
population. Dysphagia, vol. 24, no. 1, pp. 1-6, 2009. Korean adults living in local communities. International Journal of
[4] R. Manrique-Huarte, D. Calavia, A. H. Irujo, L. Girón, and M. Advanced Computer Science and Applications, vol. 9, no.10, pp. 222-
Manrique-Rodríguez, Treatment for hearing loss among the elderly: 227, 2018.
auditory outcomes and impact on quality of life. Audiology and [22] J. Chorowski, J. Wang, and J. M. Zurada, Review and performance
Neurotology, vol.21(Suppl. 1), pp. 29-35, 2016. comparison of SVM-and ELM-based classifiers. Neurocomputing, vol.
[5] M. Pigliautile, F. Chiesi, C. Primi, S. Inglese, D. Mari, D. Simoni, E. 128, pp. 507-516, 2014.
Mossello, and P.Mecocci,Validation study of the Italian version of [23] Y. C. Wu, Y. S. Lee, and J. C. Yang, Robust and efficient multiclass
Communication Activities of the Daily Living (CADL2) as an ecologic SVM models for phrase pattern recognition. Pattern recognition, vol. 41,
cognitive assessment measure in older subjects.NeurologicalSciences,e- no. 9, pp. 2874-2889, 2008.
pub:doi.org/10.1007/s10072-019-03937-w, 2019. [24] N. Herrmann, N. Mittmann, I. L. Silver, K. I. Shulman, U. A. Busto, N.
[6] P. H. Noël, J. W. Williams, J. Unützer, J. Worchel, S. Lee, J. Cornell, H. Shear, and C. A. Naranjo, A validation study of the Geriatric
W. Katon, L.H.Harpole, and E.Hunkeler, Depressionand comorbid Depression Scale short form. International Journal of Geriatric
illnessin elderly primary carepatients: impacton multiple domains of Psychiatry, vol. 11, no. 5, pp. 457-460, 1996.
397 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[25] B. Scholkopf, and A. J. Smola, Learning with kernels: support vector [31] T. Rakhshani, D. Shojaiezadeh, K. B. Lankarani, F. Rakhshani, M. H.
machines, regularization, optimization, and beyond. MIT press, Kaveh, and N. Zare, The association of health-promoting lifestyle with
Cambridge, 2001. quality of life among the Iranian elderly. Iranian Red Crescent Medical
[26] Available:https://ratsgo.github.io/machine%20learning/2017/05/29/SV Journal, vol. 16, no. 9, e-pub: doi.10.5812/ircmj.18404 2014.
M2/ [32] Y. S. Kim, and K. H. Suh, Living arrangement, satisfaction with
[27] Statistics Korea, Basic Research on Life Satisfaction of the Elderly, living, end depression among the Korean elderly. Korean
Statistics Korea, Daejeon, 2018. Journal of Gerontological Social Welfare, vol. 18, no. 1, pp.
273-290, 2002.
[28] M. H. Alshamali, M. M. Makhlouf, M. Rady, A. Selim, N. Abdel, S.
Ismail, and M. Fawaz, Quality of life and its predictors among Qatari [33] J. D. Kwon, and J. Y. Cho, A study of factors influencing the life
elderly attending primary health care centers in Qatar. Middle East satisfaction of the aged. Journal of the Korea Gerontological Society,
Journal of Family Medicine, vol. 7, no. 10, pp. 9-19, 2019. vol. 20, no. 3, pp. 61-76. 2000.
[29] K. H. Jo, and H. J. Lee, Factors related to life satisfaction in young-old, [34] H. R. Hedayati, N. Hadi, L. Mostafavi, A. Akbarzadeh, and A.
old, and oldest-old women. Journal of Korean Academy of Nursing, vol. Montazeri, Quality of life among nursing home residents compared with
39, no. 1, pp. 21-32, 2009. the elderly at home. Shiraz E-Medical Journal, vol. 15, no. 4, e-pub:
doi.10.17795/semj22718, 2014.
[30] C. W. Han, E. J. Lee, T. Iwaya, H. Kataoka, and M. Kohzuki,
Development of the Korean version of Short-Form 36-Item Health [35] M. Tajvar, M. Arab, and A. Montazeri, Determinants of health-related
Survey: health related QOL of healthy elderly people and elderly quality of life in elderly in Tehran, Iran. BMC public health, vol. 8, no.
patients in Korea. The Tohoku journal of experimental medicine, vol. 1, e-pub: doi.org/10.1186/1471-2458-8-323, 2008.
203, no. 3, pp. 189-194, 2004. [36] H. K. Kim, M. Hisata, I. Kai, and S. K. Lee, Social support exchange
and quality of life among the Korean elderly. Journal of Cross-Cultural
Gerontology, vol. 15, no. 4, pp. 331-347, 2000.
398 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This work shows a methodology of implementation The concept of “smart” devices is inherent in connectivity
and testing of the system is proposed and tested with a prototype; to obtain benefits from the information [2]. Industry
it is constructed with sensors and actuators that allow monitoring automation, and monitoring systems are the main reasons for
the behavior of the system in an environment under threats. We this technology’s success, so data networks are being unified
used an IDS / IPS as a protection tool for IoT systems, based on with production networks to achieve these benefits. It has been
Raspberry Pi and Raspbian operating system. It is described in a estimated that in 2018 there were more than 7 billion IoT
block diagram the testing method used. We implemented the devices [3]. It is estimated that by 2020 there will be more than
IDS/IPS Snort tool in an embedded platform Raspberry. It 10 billion and more than 22 billion by 2025. This worldwide
presents also the state of the art of cloud frameworks that have
trend is due to the growing demand to connect devices to the
the same objective of protecting. The main contribution is the
implemented testing method for Snort that can be used with
networks.
security rules in other applications of embedded IoT devices. With this trend, has been observed clearly that the way of
the information interchange with technology will change, but at
Keywords—Security IoT; IDS/IPS software; Pentesting tools; what cost? Due to the demand for interconnection, many IoT
smart cities; prototype Raspberry developers do not consider the security in communication for
I. INTRODUCTION many reasons: Amongst these, we have processing costs,
training and algorithm implementation.
Today, Information Technology (IT) is immersed in the use
and exploitation of telecommunications networks, to which IoT devices are considered to have many weaknesses in
more devices are connecting every day to communicate with information security since their development. The following
each other (Peer to Peer) and to a central device (client-server). are examples described below:
Internet of Things (IoT) offers wide range state of the art Passwords stored in plain text.
solutions, using sensors and/or devices, which monitor to
control certain events, giving rise to new challenges to IT Outdated firmware and not encryption.
security, since information gathered can be compromised by
this variety of components. Internet of Things (IoT) is a Video streaming without encryption.
concept defined by Kevin Ashton in 1999, which describes a Communication between devices and servers in plain
network that connects people and objects [1]. These objects, text.
right now, allow computers to have “sensors”, which facilitate
them, not only to process information but gather more Over-shared data (influence of cloud utilization).
information through these devices, allowing applications to be Development bugs in the firmware.
even more “intelligent”, since it is possible to make decisions
in real-time, based on a major quantity of information. It is Use of default passwords.
possible to assure that since the first implementation of IoT up
until date there are millions of sensors interconnected. Devices have a direct interface to the internal network,
but they can be connected to the Internet, making
increase an attack risk exponentially.
399 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In addition, it has known that hackers are exploiting these Section III describes IDS/IPS tools, in Section IV describes the
vulnerabilities with current tools and techniques of their own to methodology, Section V shows the results about this work, and
achieve that goal. One of the most recent tools to detect finally, in Section VI shows conclusions.
vulnerabilities in IoT devices is Autosploit [4], since it uses
artificial intelligence in its algorithms [5]. II. SECURITY IN IOT
The main contribution of this paper is to present a testing As a response to exposed security problems, the main
method for IDS/IPS and the comparison of its response software developers propose various strategies to guarantee
implemented on the Raspberry platform to Nmap and security in interconnection components. According to [6] the
Metasploit of network attacks. The paper is organized as expansion of IoT, it has been developing in the following
follows: Section II deals with actual security in IoT, in areas, see Table I.
TABLE. I. IOT EXPANSION AREAS BASED ON [6]
Description
IoT development areas
Elements Opportunities Challenges
- Health care.
- Consumer and Retail businesses. Technologies that promote
Ensure secure information and data
Smart life - Bank Convergence. simplifications in the lives of the
exchange.
- Security. users.
- Public services
- Intelligent vehicles
- Urban mobility.
- Intercity mobility. Real-time solutions that make
Secure interconnection and secure real-
Smart mobility - Rate management and payment mobility simpler and transport
time monitoring and activation.
solutions. reliable.
- Distribution and logistics.
-Fleet management.
- Machine learning.
-Communication between machines.
Smart solutions to optimize Keep process information safe,
- Network interconnection.
Smart manufacturing production processes, controls and interconnections between machines
-Optimization of processes.
quality. using secure protocols.
-Proactive asset management.
- Improve infrastructure integration.
400 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. MAIN FRAMEWORKS IN THE CLOUD THAT OFFER SECURITY FOR IOT
They offer a cloud platform with open and flexible - Device layer. It offers an exchange of information
services to connect securely, monitor and manage -Interconnection layer. with the devices, using languages
Microsoft Azure IoT Hub [9] IoT devices and develop applications using open - Cloud access layer. such as NodeJS, .Net, Java, Python,
source SDK (Development Kit) and multiple - Hub layer. Android, IOS and C. It establishes
protocols. Working under a SaaS scheme. - Back-End for Apps. layers of security for the connection.
It offers a PaaS scheme, which allows you to It offers connectivity with IOS,
- Software for devices.
connect IoT devices to the cloud, analyze data in Android and any device that uses
Oracle Internet of Things - Hub for access to cloud
real time to integrate it into business applications, Java, Posix C, and the RESTful
Cloud Services [10] services.
allows you to establish web services and any other protocol offers a cryptography
- Offers a wide range of SaaS.
Oracle proprietary service. scheme.
401 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The research methodology used is based on an 2 Adc1 Sensor I2C, a signal for detecting AC I
experimental and applied method, therefore the process 3 Adc2 Sensor I2C, for detecting DC I
compromises several steps, which they are described in next
4 Adc3 Sensor I2C, for detecting AC Voltage
lines. The implementation of an IDS/IPS as a security scheme
on a Raspberry Pi3B+ card, is a relatively simple process, 5 Adc4 Sensor I2C, , for detecting DC Voltage
however, it is necessary to evaluate the operation of the system, 6 BH1750FVI Sensor I2C, for detecting light
to develop adequate detection rules using Snort and Sagan, to
improve the embedded system in the management and 7 PIR Sensor Detect presence /absence, ON/OFF
monitoring the network traffic and the internal state of the 8 FZ0430 Sensor Detect DC voltage
device. 9 MCP3424 Card Analog-Digital Converter with I2C
Fig. 2 shows the methodology in block diagram used to 10 Relay 2 Card Relay 2 canals 5v
design the prototype and pentesting probes. The methodology
proposes a reviewing of the state of the art in IoT security 11 LED Bulb 127v bulb
context, and related or similar projects, then a prototype is 12 Electromagnet Actuator Opening device
implemented with the IDS/IPS tool installed for pentesting and
13 Motor Actuator 12v motor
monitoring threat behavior in these devices. At the end,
feedback is proposed to improve the prototype’s components 14 Ov5647 Actuator
Infrared night vision camera with IR
and software tools for security. sensor, 5MP
402 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
C. Pentesting
IDS/IPS operation tests were performed using Nmap and
Metasploit. With Nmap, the following instructions were
applied:
#nmap –f –sS –sV –script auth 192.168.0.9
(a)
(b)
Fig. 4. Prototype Design and Testing, a) Component Integration and b) Fig. 6. Nmap Vulnerability Results.
Sensors Testing in Raspbian with Python Language.
403 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 7. Traffic Generated and Vulnerabilities Detected by Scanning. Fig. 9. Attack on SSH Service Detected with Snort Tool.
The next step was developing the brute force attack, which //Configure Snort IPS (edit snort.conf)
consists of using a dictionary attack for breaking the password #sudo vi /usr/local/etc/snort/snort.conf
of a user account in the attacked system. Add -ipvar HOME_NET 192.168.0.0/24 –make this match your
internal network;
#Msfconsole Add -ipvar EXTERNAL_NET !$HOME_NET //IPs of network
#Search ssh home
#Use auxiliary/scanner/ssh/ssh_login Add -var RULE_PATH rules
#Show options Add -var WHITE_LIST_PATH rules //IPs from host allowed
#Set BLANK_PASSWORDS true Add -var BLACK_LIST_PATH rules
#Set PASS_FILE /root/Escritorio/pass.txt Add this to the end after “decompress_depth 65535”
#Set USER_FILE /root/Escritorio/users.txt max_gzip_mem 104857600
#Set RHOSTS 192.168.0.9 -Add this line -output unified2: filename snort.log, limit 128
#Run -delete or comment out all of the “include $RULE_PATH” lines
except:
Intrusion attempts, generated by Metasploit, detected and #include $RULE_PATH/local.rules
#include $RULE_PATH/snort.rules–add after local.rules7.
blocked by the Snort tool. //Now the following rules are uncommented, for:
Fig. 8 shows the amount of traffic generated by the attack preprocessor normalize_ip4
preprocessor normalize_tcp: ips ecn stream
trying to hack and block the web service this traffic was preprocessor normalize_icmp4
blocked by the IDS/IPS. preprocessor normalize_ip6
preprocessor normalize_icmp6.
Fig. 9 shows the amount of traffic generated by trying to
hack and compromise the ssh service, using a keys dictionary.
The metric was the traffic generated by the Metasploit tool. VI. RESULTS
As described, the trend of using IoT components in
To configure the IPS is necessary to activate two basic
industry 4.0 is perhaps the most complex challenge for
elements, the whitelist (allowed hosts) and the blacklist
security, it is changing the way that information is generated
(banned attacker hosts). Finally, add the preprocessing
and exchanged. The problem observed is that due to the
directives so that the IDS automatically applies the rules:
rapidity with which IoT devices are produced and used, due to
demand, the communication security between the components
is not properly established. The origin of the threats in the IoT
derives from lack of training, investment, staff capacity, and
security schemes.
It has demonstrated that the components, which were
integrated into an IoT system, in their wild they were weak
about security characteristics. They were not built
considerating security parameters. So we made some
penetration testing to demonstrate their behavior under some
kind of attacks. We used at the same time an IDS/IPS tools in
these tests, to demonstrate that is necessary to support and help
an IoT system. The attacks went into the system in a direct
way, only the IDS/IPS system helped to detect them. With
Nmap, we obtained a list of vulnerabilities in the prototype
Fig. 8. Result of the DDoS Attack for Web Service Detected with Snort. then we did a DDoS attack on port 80, we also used brute force
attack trying to guess a user’s password over SSH service. In
these attacks that were tested, Snort detected the unusual traffic
and behavior and it sent messages and warnings.
404 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
It is worth mentioning that initially in the development of [5] Mosca, D. “Hacking the internet of things just got easier – it’s time to
the project, the manufacturing state of the prototype look at your security”. [Online]. Available: https://www.computer
weekly.com/opinion/Hacking-the-Internet-of-Things-just-got-easier-its-
components was analyzed and the sensors and actuators were time-to-look-at-your-security. [Accessed April, 2019].
implemented. With this, it was determined that despite these [6] Rishi Rahul & Saluja Rajeev. “Future IoT”. Ernst & Young Associates
components, it was advisable to improve its security with LLP, Published in India. 2019
specialized tools; this phase of work was done with the Snort [7] Lovejoy C., Watson R. & Pizzala J.. “Internet of Things and Operating
software. The Sagan tool proved to be more demanding in the Technology Security”. [Online]. Available: https://www.ey.com/en_gl/
use of memory and processing, recommending its implementation advisory/iot-operating-technology-security. [Accessed August 2019].
in multi-threaded architectures that can support the demand. [8] Amazon Web Services, “Internet de las cosas, Plataforma como servicio
AWS IoT,” 2019. [Online]. Available: https://aws.amazon.com/es/iot/.
VII. CONCLUSION AND FUTURE WORK [Accessed: 23-Apr-2019].
[9] Microsoft. (2019). IoT Hub | Microsoft Azure. Retrieved April 23, 2019,
In this work, it has demostrated how strong or weak are the from https://azure.microsoft.com/es-mx/services/iot-hub/
IoT components against some common attacks which can be [10] Oracle, “Internet of Things | Oracle Cloud,” 2019. [Online]. Available:
found on the Internet. We found that the components of an IoT https://cloud.oracle.com/iot. [Accessed: 23-Apr-2019].
system were built with no considerations on security schemas [11] IBM, “IBM Watson Internet of Things (IoT)”, 2019. [Online].
and response against common attacks. So, is recommend Available: https://www.ibm.com/mx-es/internet-of-things. [Accessed:
installing an IDS/IPS to secure an IoT system, to prevent and 23-Apr-2019]
warning against some cyber attacks. [12] G. I. Xively, “IoT Platform for Connected Devices”, 2019. [Online].
Available: https://xively.com/. [Accessed: 23-Apr-2019].
The implementation of Snort as an intruder detection [13] Samsung Co., “IoT Cloud Platform, Samsung ARTIK cloud services”,
system allowed real-time detection of port scanning and 2019. [Online]. Available: https://artik.cloud/. [Accessed: 02-May-
attempts to breach the system from other hosts; providing an 2019].
opportunity to measure how systems are compromised. This [14] Altair Engineering Inc., “Altair SmartWorks” 2019. [Online]. Available:
provides a new opportunity for an investigation to model this https://www.altairsmartworks.com/index.php/. [Accessed: 02-May-
2019].
behavior. Finally, it is important considering the stabilization
of rules for IDS/IPS so that the system permits secure [15] Adafruit, “Welcome to Adafruit IO” 2019. [Online]. Available:
https://io.adafruit.com/. [Accessed: 02-May-2019].
communication without repudiation.
[16] Ubidots, “IoT platform Ubidots” 2019. [Online]. Available:
Our future works can treat over new penetration tests, set https://ubidots.com/. [Accessed: 02-May-2019].
up new rules in the IDS/IPS, encrypted messages among [17] MyDevices, “The IoT Solutions Company” 2019. [Online]. Available:
wireless components and integrate all components and tools to https://mydevices.com/. [Accessed: 02-May-2019].
implement a platform of IoT secure scheme. [18] Macchina.io, “IoT Edge Device Software Development and Secure
Remote Access Solutions”, 2019. [Online]. Available:
ACKNOWLEDGMENTS https://macchina.io/. [Accessed: 02-May-2019].
[19] ThingSpeak, “IoT Analytics, ThingSpeak Internet of Things”, 2019.
We thank Tecnológico Nacional de México/Instituto [Online]. Available: https://thingspeak.com/. [Accessed: 02-May-2019].
Tecnológico de Morelia, as well as Universidad Nacional [20] Arduino, “Arduino” 2019. [Online]. Available:
Autónoma de México, for the support granted for this research, https://www.arduino.cc/en/IoT/HomePage. [Accessed: 02-May-2019].
together with project 5774.19-P “Development and [21] Tomas Zitta, “Penetration Testing of Intrusion Detection and Prevention
implementation of a secure IoT architecture, based on System in Low-Performance Embedded IoT Device” (2018). IEEE-
penetration tests, using fuzz models and detection systems with Xplore. Retrieve: https://ieeexplore.ieee.org/document/8624734.
[Accessed: July-2019].
an IDS and IPS” Spanish translate “Desarrollo e
[22] Cisco Systems, “Snort - Network Intrusion Detection & Prevention
implementación de una arquitectura IoT segura, basado en System.” [Online]. Available: https://snort.org/. [Accessed: 13-Jul-
pruebas de penetración, utilizando modelos fuzz y sistemas de 2019].
detección con un IDS e IPS”. We also thank professors Abel A. [23] Project Suricata, “Suricata, Open Source IDS/IPS/NSM engine.”
Pintor from ITMorelia and from IIES-UNAM who helped the [Online]. Available: https://suricata-ids.org/. [Accessed: 13-Jul-2019].
development of the prototype, Diego Cabrer, Atzimba López, [24] The Zeek Network Security Monitor, “The Zeek Network Security
Alberto Valencia, and Yumi Tzib. Monitor.” [Online]. Available: https://www.zeek.org/index.html.
[Accessed: 13-Jul-2019].
REFERENCES
[25] OSSEC Project Team, “OSSEC -World’s Most Widely Used Host
[1] Haroon A., Naeem W., Shah M. A., Kamram M., Asim Y. & Javaid Q. Intrusion Detection System-” [Online]. Available:
“Constraints in the IoT: The World in 2020 and Beyond”. International https://www.ossec.net/. [Accessed: 13-Jul-2019].
Journal of Advanced Computer Science and Applications,Vol. 7, No. 11,
2016. [26] Tripwire, “Cybersecurity and Compliance Solutions”. [Online].
Available: https://www.tripwire.com/. [Accessed: 14-Jul-2019].
[2] Santoso F. K. & Vun N. C. H. “Securing IoT for smart home system”.
International Symposium on Consumer Electronics (ISCE), 2015. IEEE. [27] Linux, “Intrusion detection with AIDE”. [Online]. Available:
ISBN: 978-1-4673-7365-4. DOI: 10.1109/ISCE.2015.7177843. https://www.linux.com/news/intrusion-detection-aide. [Accessed: 14-
Jul-2019].
[3] Qinghe D., Houbing S. & Xuejie Z. “Social-Feature Enabled
Communications Among Devices Toward the Smart IoT Community”. [28] Samhain design labs, “Samhain Labs” [Online]. Available: https://www
IEEE Communications Magazine. Volume 57 Issue 1. January-2019. .la-samhna.de/samhain/index.html. [Accessed: 18-Jul-2019].
DOI: 10.1109/MCOM.2018.1700563. [29] Fail2ban Project, “Fail2ban.” [Online]. Available: https://www.fail2ban
[4] Rouhiainen Tuukka. “Scanning the Internet to find security loopholes”. .org/wiki/index.php/Main_Page. [Accessed: 19-Jul-2019].
Proceedings of the Seminar in Computer Science: Internet, Data and [30] Q. I. S. Sagan Project, “The Sagan Log Analysis Engine | Quadrant
Things (CS-E4000). Computer Science at Aalto University. 2018. Information Security.” [Online]. Available: https://quadrantsec.com/
sagan_log_analysis_engine/. [Accessed: 19-Jul-2019].
405 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The scale and frequency of sophisticated attacks Many methods are used to block DDoS attacks, including
through denial of distributed service (DDoS) are still growing. some:
The urgency is required because with the new emerging
paradigms of the Internet of Things (IoT) and Cloud Computing, The signature-based approach: it requires an a priori
billions of unsecured connected objects will be available. This knowledge of the elements related to the signature of
document deals with the detection, and correction of DDoS attacks, see SNORT [2]. Signatures are manually built
attacks based on real-time behavioral analysis of traffic. This by security experts. The authors of [3] analyze previous
method is based on Software Defined Network (SDN) attacks to look for a match with incoming traffic to
technologies, Bloom filter and automatic behaviour learning. detect intrusions. Signature-based techniques are only
Indeed, distributed denial of service attacks (DDoS) are difficult effective in detecting the traffic of known DDoS
to detect in real time. In particular, it concerns the distinction attacks; while new attacks or even slight variations of
between legitimate and illegitimate packages. Our approach old attack go unnoticed.
outlines a supervised classification method based on Machine
Learning that identifies malicious and normal packets. Thus, we Anomaly-based detection: the anomaly-based system
design and implement Defined (IDS) with a great precision. The uses a different method. It treats any network
results of the evaluation suggest that our proposal is timely and connection that violates the normal profile as an
detects several abnormal DDoS-based cyber-attack behaviours. anomaly. The anomaly is revealed if incoming traffic
deviates significantly from normal profiles, see [4] and
Keywords—Distributed denial of service; intrusion detection [5]. To detect DDoS attacks, it is first necessary to
software; software defined network; machine learning; know the overall normal behaviour of the system traffic
synchronize; acknowledgment; bloom filter and then to find deviations from this behaviour. The
I. INTRODUCTION anomaly-based technique can detect new attacks.
However, it can initiate many false alarms.
Over the past decade, DDoS attacks have been a powerful
threat to the security of many Internet service providers, and Packet filtering: packets entering and leaving the
have resulted in economic losses for them. DoS attacks cause a network protect the network against attacks from any
denial of service to legitimate requests by depleting network source. This technique uses server firewalls, router
resources and services. To maximize impact, the attack will be based packet filtering [6]. This requires the installation
launched from distributed sources, called attacks through of filter input and output packets on all routers. It is
denial of distributed service. In most cases, these attacks are used to filter the spoofed IP address, but approaches to
launched by botnets. The largest DDoS attack on the latest prevent it need a global implementation that is not
records occurred in February 2018 as revealed by the Git Hub. practical [7].
The attack came from more than thousand different European In this article, we set up IDS capable of detecting anomalies
Union countries out of tens of thousands of single endpoints. based on Machine Learning techniques. The volume of data to
This was the one amplification attack using Memcached be studied is enormous, so we use SDN technology for
technology that peaked at 1.35Tbps. Another major DDoS efficient data processing. We also used the Bloom filter, which
attack is the Mirai [1] botnet attack that was used in a high is a probabilistic structure for storing and accessing data
volumetric DDoS of about 1.1 Tbps that destroyed a large part efficiently. This document is structured as follows: Section 2
of Dyn's database in October 2016. Mirai has successfully describes some approaches used to solve DDoS attack
ordered nearly 100,000 robots by exploiting the low security of problems. Section 3 outlines our method of resolution and then
cameras, home routers, digital recorders and printers with Section 4 illustrates the results and discussion. Finally, the
default credentials used for their telnet ports. conclusion is presented in Section 5.
406 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
407 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A. SDN Architecture The Bloom filter stores in the form of a table of bits that
The data plan and the control plan are increased tenfold. represent the IP addresses considered malicious, see Fig. 3.
Thus, the administrator defines the rules in the controller, and Consider ℱ = {ip1, ip2, ..., ipn}, the n IP addresses that
they are instantly transmitted in the network equipment. describe the array of n bits. Initially all bits are at 0.
Fig. 2 illustrates the SDN architecture, which consists of Let ℋ = {h1, h2, ..., hk}, all the independent hash functions
three layers. The lowest layer is the infrastructure layer, also stored in p.
called the data plan. It includes the elements of the transfer
network. The responsibilities of the routing plan are mainly For each ipx on ℱ :
data transfer, as well as monitoring of local data transmission,
hj(ipx) =1 for 1 ≤ j ≤ k. (2)
information and statistical collection.
The layer above is the control layer, also called the control To check if an attack suspect ipx address is in ℱ. We check
plan. He is responsible for the programming and management that all hj(ipx) = 1, otherwise h(ipx) = 0 is not malicious. This
of the routing plan. To this end, it uses the information process can generate false positives. In other words, it can
provided by the transmission plan and defines the operation happen that for an ipx address we have hj(ipx) =1 while it is not
and routing of the network. It includes one or more controllers malicious.
that communicate with the elements of the transmission In our approach, false positives are negligible because the
network through standardized interfaces, known as southbound probability of their existence is low. Indeed, let us consider m,
interfaces. the size of the Bloom filter, n the number of hash functions. Let
The application layer contains network applications that X be a random variable representing all the bits. Thus, the false
can introduce new network functionality through APIs, such as positive rate can be evaluated by:
security and management, transfer schemes or control layer 1 nk
support in network configuration. It has an abstract and global P ( X = 0) = (1 - )
view of the network from the controllers and uses this m (3)
information to provide appropriate advice to the control layer. In [17] have shown that this rate is very low because:
The interface between the application layer and the control is
called the northward interface. p
k = ln(2) (4)
Northbound APIs can be used to facilitate innovation and n (4)
enable efficient network orchestration and automation to align
with the needs of different applications through SDN network When k=10 and p=20n, the probability of a false positive is
programmability. We will use this property of the application 0.0000889. This result justifies the use of the Bloom filter in
layer to implement a TCP flooding attack detection module. the detection of DDoS attacks based IDS architecture.
408 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In our approach, we have combined the advantages of the b) The control plan: After receiving a new packet, the
behavioural filter and Machine Learning as shown in Fig. 4. controller processes, calculates and creates a new flow entry,
In our method, network traffic must be collected from which it then sends to the switch. The switch receives the
switches and then used to build the drive and classification set. message from the controller, adds the new entry to its flow
The management of packets entering the network is presented table, and manages the packet as defined in the entry [18].
in Fig. 4. When a new packet arrives at the switch, if it belongs When the packet is unknown to the controller, the Openflow
to an existing flow in the flow table, it updates the flow protocol sends the packet header to the Identity Attack and
statistics otherwise, a "Packet-In" message is sent to the Storage Detection (ADIS) module in the application plan.
Openflow controller. The controller responds with a "Packet- c) The application plan: The ADIS module of the
out" message on the attitude to be followed according to the application layer is designed to analyze the SDN network flow
pre-established rules. tables and collect traffic flows by inspecting the IP header
Switches in the data plan uses tables to route packets. This {src_ip, src_port, dst_ip, dst_port, protocol}. Each flow can be
is possible by using entries in flow tables and a packet represented by a set of statistical characteristics, such as
processing process. According to [17] an entry in the flow table DurationSeconds, packetCount and byteCount, etc. The ADIS
consists of seven fields: Match Fields, Priority, Counters, module checks if the IP address of the packet is stored in the
Instructions, Timeouts, Cookie, Flags. Bloom filter (attacker database). Failing this, a deep analysis
based on the number of packets sent per second by the source
The Counters field allows you to know the total number of
to the Openflow switch classifies the category of the source IP
packets processed for an entry. Counters can be maintained for
address. In the following section, we propose a classification
each flow table, number of packets or bytes, flow entry, port,
queue, duration during which the entry was activated. algorithm.
2) Data classification algorithm: In order to detect the
C. ADIS: DDoS Attack Detection Module DDoS attack, the IDS must be supplied with traffic
In this section, we describe the mechanism for detecting information related to the following parameters: src_ip,
and preventing attacks. src_port, dst_ip, dst_ip, dst_port, protocol, DurationSeconds,
1) DDoS defense architecture packetCount and byteCount, etc. It uses a Machine Learning
a) The data plan: When a new packet arrives at the (ML) classification model to detect attack activity. In our
switch, it checks whether the packet header matches an entry in example, we will use the following models : linear
its flow table. discriminant analysis (LDA), k-nearest neighbors (KNN) and
If it finds an entry, it processes the packet as defined in the Support vector machine (SVM).
corresponding entry. Otherwise, he forwards the packet to the These models can learn the pattern with few training
controller in order to receive instructions after a thorough samples and produce an accurate classification by reducing
investigation. false positives.
409 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
410 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
5,00%
Proceedings of the ACM SIGCOMM 2001, Conference on Applications,
4,00% Technologies, Architectures, and Protocols for Computer
4,00%
3,00% Communications, ACM Press, New York (2001), pp 15–26.
2,00% [7] T. Peng, C. Leckie, K. Ramamohanarao, „„Protection from distributed
1,00%
denial of service attack using history-based IP filtering,‟‟ In:
1,29%
Proceedings of IEEE International Conference on Communications (ICC
0,00%
LDA KNN SVM 2003), Anchorage, AL, vol. 1, pp. 482–486.
Machine Learning Models [8] E. Fenil, P. Mohan Kumar , „„Survey on DDoS defense mechanisms,‟‟
wiley, wileyonlinelibrary.com/journal/, Décembre 2018.
Fig. 11. Average Model Error.
[9] V. Chidri, V. Balasubramani, S. Sadath Ali, S. Shrikrishna Hegde, P.
Sadanand, „„A Survey on Distributed Denial-of-service Attacks and
The performance of the model will be measured by the Defense Mechanisms,‟‟ JETIR1504089 Journal of Emerging
average misclassification error (mmce.test.mean) of each Technologies and Innovative Research (JETIR) www.jetir.org, avril
model LDA, KNN and SVM, see Fig. 11. 2015.
411 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[10] T. Jog, M. Natu, S. Shelke, „„ Distributed capabilities-based DDoS computer infrastructures,‟‟ Future Generation Computer Systems 91(1),
defense,’’ 2015 International Conference on Pervasive Computing 144{156 (2019)., https://doi.org/10.1016/j.future.2018.08.011,
(ICPC), janvier 2015, pp. 1–6. https://doi.org/10.1016/j.future.2018.08.011.
[11] C. Buragohain, M. Kalita Santosh Singh, D. Bhattacharyya, „„Anomaly [15] R. Patgiri, S. Nayak, and S. K. Borgohain, “Preventing DDoS using
based DDoS Attack Detection,‟‟ Chaitanya Buragohain, Manash Jyoti bloom filter: A survey,” ICST Transactions on Scalable Information
Kalita Santosh Singh, Dhruba K.Bhattacharyya, International Journal of Systems, vol. 5, no. 19, Article ID 155865, 2018.
Computer Applications (0975 –8887) Volume 123–No.17, August, [16] C. Tseung, K. Chow, and X. Zhang. „„Anti-DDoS technique using self-
2015, pp35-40. learning bloom filter,‟‟ In Intelligence and Security Informatics (ISI),
[12] A. Cardigliano, L.Deri et T. Lundstrom, „„Commoditising DDoS 2017 IEEE International Conference on, pages 204–204. IEEE, 2017.
mitigation,‟‟ septembre 2016, p. 523–5282016 International Wireless [17] P. Cao, „„Bloom filters-the math,‟‟ University of Wisconsin-Madisson,
Communications and Mobile Computing Conference (IWCMC). Madisson (1998). http://pages.cs.wisc.edu/~cao/papers/summary-
[13] N. Lu, S. Su, M. Jing, and J. Han, „„A router-based packet filtering cache/node8.html.
scheme for defending against dos attacks. China Communications, [18] Open Networking Foundation, „„OpenFlow Switch Specification,‟‟
11(10, 2014 ), pp136–146. Version 1.5.1 ( Protocol version 0x06), https://www.opennetworking.org
[14] R. Koning, B. de Graaff, G. Polevoy, R. Meijer, C. de Laat, P. Grosso, /images//openflow-switch-v1.5.1.pdf.
„„Measuring the efficiency of SDN mitigations against attacks on
412 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The aim of this paper is the design, simulation, The electronic design and motion planning of a robot
construction and programming of the autonomous vehicle, based on decision making regarding its straight motion and
capable of obstacle avoidance, object tracking also image and precise turn using Artificial Neural Network (ANN) has been
video processing. The vehicle will use a built-in camera for proposed in [3, 8]. The ANN helps in learning of robot so that
evaluating and navigating the terrain, a six-axis accelerometer it performs motion autonomously. The weights calculated are
and gyro for calculating angular velocities and accelerations, implemented in microcontroller [3].
Arduino for interfacing with motors as well as with Raspberry Pi
which is the main on-board computer. The design of the vehicle is Obstacle avoiding technique is very useful in real life, by
performed in Autodesk Fusion 360. Most of the mechanical parts changing the IR sensor by a kinetic sensor, which is on type of
have been 3D printed. In order to control the chassis of the microwave sensor whose sensing range is very high and the
vehicle through the microcontrollers, the development of the output of this sensor vary in according to the object position
PCB was required. On top of this, a camera has been added to changes [4].
the vehicle, in order to achieve obstacle avoidance and perform
object tracking. The video processing required to achieve these The purpose of this paper is to build a vehicle that will be
goals is done by using OpenCV and Convolutional Neural driven autonomously through decisions taken by Artificial
Network. Among other objectives of this paper is the detection of Intelligence. In more detail, the objectives are:
traffic signs. The application of the Convolutional Neural
Network algorithm after some of the examinations made has Design of the vehicle, and its construction;
shown greater precision in recognizing STOP traffic sign of Design and development of the PCB board that will
different positions and occlusion ratios, and finding the path for control the vehicle;
the fastest time.
Development of a communication method between the
Keywords—Image processing; traffic sign; object tracking; vehicle and the laptop;
autonomous vehicle; convolutional neural network
Development of the electronic circuits and program for
I. INTRODUCTION serial communication between Arduino and Raspberry
In recent years, autonomous vehicles have become of great Pi;
interest to the research and industrial communities. The Creating a server for transmitting video and sensors
literature related to autonomous vehicles takes mainly two data from Raspberry Pi to the laptop;
directions hardware and software developments. For the full
functionality of autonomous vehicles, there is no need for Detection of traffic signs;
more hardware. Software and testing is where much work
needs to be done. Many researchers have been oriented in Stopping the vehicle if there are any obstacles ahead;
designing the small vehicle prototypes due to cost Building a Convolutional Neural Networks (CNN)
effectiveness and reduction of software testing with different model for predicting the movement.
sensors and algorithms.
Aim of this paper is to get closer in trend with today's
The Raspberry Pi as a processing chip has been used to companies that produce autonomous cars, one of which is well
build a monocular vision autonomous car prototype. An HD known, Tesla. But there are also many companies that are
camera along with an ultrasonic sensor has been used to working to release autonomous cars on our roads.
provide necessary data from the real world to the car [1].
The paper is organized in 8 sections, starting with
A method for autonomous control and decision making mechanical design, electric circuits, controllers programming,
and reporting system, the types of mini robots contains self- training the models for the detection of traffic signs,
neural schema framework for autonomous control has been programming and control with CNN, data processing method,
proposed by [2]. and, concluding remarks with future work.
*Corresponding Author.
413 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
II. MECHANICAL DESIGN AND 3D MODELLING BY USING converter for Arduinos, 2x Arduino Pro Mini, buzzer, power
THE AUTODESK FUSION 360 amplifier for RGB lights, voltage divider, and distances for
Raspberry Pi [5].
For modelling and designing of the parts, the Autodesk
software, Fusion 360, has been used. Initially, motors and It’s worth to mention that the L298N motor drivers are not
wheels have been designed by measuring the actual physical placed on the board but are connected to it, due to their large
parts. Then, the assembly of the motors with wheels and the dimensions.
respective joints have been done. The chassis has been
designed in seven different parts, because the printing area of A. Development of the Transmitter Box
3D printer used was 200x200 mm, and the dimensions of the For laptop and Arduino to communicate, the chosen
model exceed this area. Following this step, the motors and method was through the Radio Modules, in this case through
wheels have been assembled into chassis. Other parts needed NRF24L01. For the implementation of the transmission box,
for the implementation of this project have been designed, and Fig. 3, a 3D box has been designed and printed initially, and
are shown below on Fig. 1. an Arduino Pro Mini, and a NRF24L01 radio module have
been installed.
Since the Arduino Pro Mini could not be programmed
directly from the laptop, it was necessary to use an FTDI.
The NRFL01+ module with antenna is capable for
providing communication up to 1km in open terrain.
414 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
3.0 V Cutoff
C. Batteries Discharge: 500 mA Standard
Given the huge power draw from all four DC motors and 3750 mA Maximum
the large number of electronic components, only two types of
batteries were suitable for high discharge rates: Li-Po and Li-
Ion. The latter are chosen since they are safer, lighter, and are
cheaper and also have a smaller size. The type of batteries
used is LGDBB31685, Table I. These batteries are recycled
from laptop batteries. Table I shows the specifications of this
type of batteries.
Fig. 5. Sensors on the Rear Side of the Autonomous Vehicle. Fig. 7. Top View of the Autonomous Vehicle Model.
415 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 8. Method of the PWM Function. Fig. 11. Voltage Divider Scheme.
416 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
417 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
0 DEGREE 5 10 20 30
DEGREES DEGREES DEGREES DEGREES
Confidence
Fig. 15. Python Scripts Execution through Putty. Fig. 18. Performance with Slang Angles Parallel to Sign.
418 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
240 SSD
0.547259 Laptop Intel core i3
@2.53 2.53
6 GB +750 GB
GHz
HDD
Raspberry Pi ARM cortex 8 Gb SD
@1.2 GHz 1 GB
3B A53 card
0.0385611
5% 8% 10% 15% 50%
Confidence
VIII. CONCLUSIONS
For this research paper, mechanics was applied for
designing the vehicle and mounting of the entire hardware,
electronics was used to develop a circuit board, and the
hardware was synchronized with the electrical circuits through
the programming language. The fusion of all of these has
resulted into a mechatronic project.
Fig. 20. Data Collection Process for CNN Training. In conclusion, a vehicle capable of moving in all the
necessary directions (forward, forward right, forward left,
The first step in CNN training is manually driving the backwards, backwards right, backwards left as well as turn
vehicle. In the Python code, the vehicle is driven by the W, A, and rotate in place) was “born”.
S and D keys: W for forward, S for reverse movement, D-
right, A - left, WD - right front, WA - forward left, SD- The vehicle has been equipped with sensors for gathering
backwards right as well as SA - backwards left. During necessary information about the surrounding environment as
manual data collection, Python scripts for video transmission well as camera to collect the pictures of the traffic signs.
from Raspberry Pi on the laptop should be executed at the Based on the CNN algorithm one can conclude that the
same time, so that for each key pressed, the laptop will capture influence due to occlusion is much larger than that of slant
the frame and the corresponding key. These frames, as angle. Therefore, it can identify the stop sign even if the sign
mentioned earlier from a 2D matrix, are transformed into a is inclined 30 degrees and with 15% occlusion.
string and stored as a *.npz file. Data collection process is
very important because the accuracy of the CNN training and The importance of this paper can be seen in the fact that
its prediction depends directly on the data collection process. some steps have already been made to navigate the
In this case, over 440 Mbytes of data for CNN training were autonomous car model that can be compared with the
collected as in Fig. 20. autonomous machines of the elite companies in this field.
VII. DATA PROCESSING METHOD There is also room for improvements that will be made in
future models of this vehicle using other learning methods for
Throughout the paper is mentioned that the video and data traffic sing recognition.
are transmitted from the RPi to the laptop. Knowing the fact
that RPi is a single board computer, the question why the REFERENCES
processing isn’t done directly on Raspberry Pi instead of [1] Pannu G. S., Ansari, M. D. & Gupta P. Design and Implementation of
laptop might arise. This all comes down to processing power. Autonomous Cars Using Raspberry Pi, International Journal of
Computer Applications (0975 – 8887), pp.222-222, Volume 113, No.9,
Below is given a Table II showing key differences between March 2015.
these two. From the table, it’s clear that laptop’s processing [2] M.Karthikeyan, Mr. G.Sreeram, M.Tech, (Ph.D). Intelligent Exploration
power is far “superior” compared to Raspberry Pi’s. and Surveillance Robot In Defense Environment. International Journal
of Advanced Research in Electrical, Electronics and Instrumentation
The way how the entire hardware “talks” to each other is Engineering. Vol. 3, Special Issue 1, February 2014.
shown in Fig. 21.
419 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[3] G. N. Tripathi and V.Rihani. Motion Planning of an Autonomous networks. International Journal of Applied Pattern Recognition, 5(3),
Mobile Robot using Artificial Neural Network. Mody Institute of 240-250.
Technology and Science, Lakshamangarh, Sikar, Rajasthan. [8] Hwu, T., Isbell, J., Oros, N., & Krichmar, J. (2017, May). A self-driving
[4] Rakesh Chandra Kumar et al. Obstacle Avoiding Robot – A promising robot using deep convolutional neural networks on neuromorphic
One. International Journal of Advanced Research in Electrical, hardware. In 2017 International Joint Conference on Neural Networks
Electronics and Instrumentation Engineering Vol. 2, Issue 4, April (IJCNN) (pp. 635-641). IEEE.
2013. [9] Ishibushi, S., Taniguchi, A., Takano, T., Hagiwara, Y., & Taniguchi, T.
[5] Beqa, F. Development of autonomous vehicle driven by Artificial (2015, November). Statistical localization exploiting convolutional
Intelligence, University of Prishtina, 2018. neural network for an autonomous vehicle. In IECON 2015-41st Annual
[6] Pajaziti, A., & Bajrami Xh. & Paliqi A. Path Control of Quadruped Conference of the IEEE Industrial Electronics Society (pp. 001369-
Robot through Convolutional Neural Networks, 18th IFAC Conference 001375). IEEE.
on Technology, Culture and International Stability, Sept 13-15, 2018, [10] Tai, L., Li, S., & Liu, M. (2017). Autonomous exploration of mobile
Baku, Azerbaidschan, 2018. robots through deep neural networks. International Journal of Advanced
[7] Bajrami, X., Gashi, B., & Murturi, I. (2018). Face recognition Robotic Systems, 14(4), 1729881417703571.
performance using linear discriminant analysis and deep neural
420 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—A noticeable objective of this work is to experiment in order to get the comparison result for both methods. Here,
and test an optimization problem through comparing hill- the proposed hybrid method conjoining two methods LHC and
climbing method with a hybrid method combining hill-climbing HC is a novel-configuration in optimizing. It is an optimization
and Latin-hyper-cube. These two methods are going to be tested method by dividing search-area via several definitive-clusters,
operating the same data-set in order to get the comparison result and conducting searching alternatives in each cluster.
for both methods. The result shows that the hybrid model has a
better performance than hill-climbing. Based on the number of The organization of this paper is as follows. Section II
global optimum value occurrence, the hybrid model presents a literature overview about optimization, HC, and
outperformed 7.6% better than hill-climbing, and produced LHC. The related works and experimental method is going to
more stable average global optimum value. However, the model be delivered in Section III and Section IV, respectively. In
has a little longer running time due to a genuine characteristic of Section V, we present experimental result and analysis. Finally,
the model itself. Section VI concludes the paper.
Keywords—Hill-Climbing; Latin-Hyper-Cube; Optimization II. LITERATURE OVERVIEW
I. INTRODUCTION A. Optimization
Optimization, also known as mathematical programming, is Optimization derived from Latin words „optimus‟ which
a process or action to obtain the highest achievable has the „best‟ meaning, it can be interpreted as a process of
performance under the given constraints. The final goal of all finding conditions that provide optimal value from an objective
optimization is to minimize the effort required or to maximize function [12]. The optimal value can be a minimum value or a
the desired benefit. Several methods for optimization have maximum value of the objective function in accordance with
been developed are hill-climbing (HC) [1], simulated- the existing problems (Fig 1).
annealing (SA) [2], generic algorithm [3], particle swarm The combination method L2HC will be discussed in this
optimization [4], ant colony optimization [5], and others. experimental paper. The HC method itself is included in the
Many researchers have performed various studies in the method of mathematical programming or modern (non-
scope of optimization. Author in [6] conducted a study to traditional) optimization techniques that are very useful for
improve the sampling algorithm in the Bayesian network by finding the optimal value of an objective function of several
applying the Latin-hyper-cube (LHC) sample. This study variables that are in certain constraints. They are heuristic
showed a better result compared to applying simple random methods and are often faster than exact methods, especially in
sampling. Author in [7] discussed the Quasi-Newton methods the case of non-linear objective functions with many variables.
which are an iterative optimization method. There is also a While the LHC method is included in the statistical sampling
study that used the hill-climbing optimization method which method which is a method for analyzing experimental data and
was found to effectively detect grids on microarray images developing empirical models to get the most accurate
taken from databases from GEO and Stanford genomic representation of the physical situation [13].
laboratories [8]. Also, [9] constructed an optimization model B. Hill Climbing
using simulated-annealing and hill-climbing then compared
them. The result said that hill-climbing consumed the shortest HC is a method that aims to find local maximum or
running time. Author in [10] applied genetic algorithm to minimum values through simple iterations that continue to
optimize the control (process) parameters. Finally, [11] move towards increasing values if looking for maximum
explained the success of simulated-annealing method for global values or decreasing values if looking for minimum values
optimization problems by studying the ideal version of the from an objective function to finding the nearest peak value or
algorithm. closest valley point [12]. This method is very simple and has
been successfully applied to various optimization problems.
The objective of this study is to scientifically experiment an The success of this method is due to the fact that choosing a
optimization problem by comparing HC method with a hybrid heuristic that more accurately predicts the actual solution that
method combining LHC and HC (L2HC). While HC is going produces more opportunities to obtain the optimal solution
to use a random starting point in any position, the hybrid [1][8][14][15]. Pseudocode of HC method is displayed in Code
method is divided starting point into several identified clusters. 1. Searching for optimum value with HC method is greedy and
This two methods are going to be tested using the same data set the search starts from a random starting point. This causes the
421 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
optimum results can be the local optimum, except for the Code 2 shows a pseudocode for the LHC method when
random value of the luckiest starting point [16]. implemented as an optimization method. Here, randomizing a
position from the population is required, and the clusters
checked then by using a function named . It
will return a Boolean value and refers to equal status
( ) variable, returned „true‟ if it exceeds the maximum
number in that group, and „false‟ if it able. This random
process will stop when is „false‟.
Code 2. Pseudocode of Latin-Hyper-Cube Method [12]
Procedure LatinHypercubeSampling()
Begin
<...variables definition...>
<...parameters cluster making...>
bestVal <-- 0
While(loopCount)
//randomizing new parameter combination and
//checking the equality
Fig. 1. Illustration of Optimization for Maximum and Minimum Value [13]. eqStatus <-- true
While(eqStatus=true)
There are several parameters needed for the HC method, //randomizing new parameter
some of them are mandatory; such as, , , cPos <-- random()
//checking cluster equality status
, which respectively represents the current value, eqStatus<--checkCluster(cPos)
neighbor's best value, and best value (local or global optimum). End while
Based on the pseudocode, it can be comprehended that the cVal <-- objFunction(cPos)
//checking the best value
iteration process is going to be terminated when the local If(cVal > bestVal)
optimum value has been obtained. That is when there is no bestVal <-- cVal
value from the variable that is better than the bestPos <-- cPos
value of the variable. End if
loopCount++
Code 1. Pseudocode of Hill Climbing Method [12] End while
End
-
Procedure HillClimbing()
Begin
<...variables definition...> III. RELATED WORKS
//randomizing new parameter combination
cPos <-- random() Numerous studies exploring about LHC or HC method
cVal <-- objFunction(cPos) have been already conducted by many researchers. Author in
//looping until local optimum is found [20] adopted LHC sampling to design a sophisticated gas
While(search is not terminated)
//finding the best neighbor
turbine. This sampling method is applied recursively to identify
bestNeighbor <-- getBestNeighbor() the most important input parameters. Author in [21] proposed a
//finding local optimum new method named „IDLHCSA‟ for history matching to get
If(cVal>=bestNeighbor) reliable forecast. „IDLHCSA‟ combines iterative discrete
Local optimum is found
Search is terminated Latin-hyper-cube (IDLHC) to find good matched models with
Else //move to next position SA method.
cVal <-- bestNeighbor
cPos <-- bestNPos Furthermore, [22] examined the performance of HC
End if method for mesh router node placement in wireless mesh
End while
End
network. The result shows that connectivity and user coverage
are achieved well. Author in [23] applied smart HC using the
C. Latin-Hyper-Cube ideas of LHC sampling to find an optimal configuration for
LHC was proposed by [17], which is a method for selecting web application server. This proposed method can learn from
samples from populations. LHC has been proven to be able to previous searches and more efficient than traditional heuristic
reduce variance [18] and produce better analysis results methods. Also, [24] operated HC method to prove about
compared to simple random sampling method [19]. In this statement of „optimization has a better solution when it closer
method, how many intervals of equal probability so input to the local optimum value‟ is wrong. Besides the local
variable value can be divided into several cubes according to optimum value, number of steps to reach the local optimum
the intervals are firstly determined. Each parameter also needs to be considered.
combination is randomly selected and is determined how many
combinations of parameters can be in the same group. After Particular in hybrid model, [25], in discussing a
getting a combination of parameters randomly, check which communication behavior for social dynamical systems, studied
group of parameters the combination is. If it exceeds the a hybrid opinion network containing of continuous valued and
maximum number of parameter combinations in that group, the discrete valued agents. The agents talked about were copier,
value must be randomized again [12]. voter, and averager agents. Here, the communication
topologies were modeled. The study concluded that the voters‟
422 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
existence has dissimilar impact on the evolution and consensus V. RESULT AND ANALYSIS
value of negotiation process. Schematically, the constructed model is configured by Fig
Additionally, [26] also investigated resilient consensus 4. It consists of three classes, where class L2HC (the class
problem in hybrid multi-agent system. The hybrid multi agent symbolizes the proposed hybrid model) is consisting two
system itself consists of continuous time and discrete time classes LHC and HC. All attributes defined in a main class,
dynamical agents. Author in [26] successfully constructed a where they practically belong to and able to be equipped by
hybrid censoring strategy to reach resilient consensus. The both classes LHC and HC.
consensus here defined as compromise between cooperative The pseudocode of L2HC is provided in Code 3. HC has
agents and Byzantine agents (as uncooperative agents). seven parameters that are initiated and defined (section B). The
IV. RESEARCH METHODOLOGY first is which defines the initial position that
will be used in the optimization search and is
This experiment involved five stages (Fig 2); i.e. the value of that .
preliminary study, construct model, data preparation,
experiment, and evaluation. In the fisrt stage, optimization, Then, defines the position of the
HC, and LHC methods were learned deeply. Then, the model best neighbor around and
was constructed using class diagram. For the next two stages, is the value of that
python 3.6 is functioned methodically for generating the data . is found when there is
set and test the optimization method. We performed this no neighbor that have better value than current value, and
experiment by using 7th generation of Intel core i7 processor, is a Boolean which indicates when the
8GB RAM, and operating system Windows 10 64-bit as the search must stop. There are also operations that are used here,
hardware and software specifications. With the help of they are where the objective function for
interpolate.lagrange from scipy library and linspace from optimization is defined, for initial
numpy, data are generated (Fig. 3) and the equation is provided positioning, to find the best neighbour
in 1). It is going to produce interpolated values randomly. value, , and to find the best
The last stage is to perform an evaluation by analyzing the value.
experiment result.
The idea of hybrid L2HC method is applying LHC method
In addition, while doing the experiment stage, each method to HC method. This method is done by dividing the data into
randomizes 100 starting point for one process. It means that several clusters using clustering operation and running a HC
HC will do 100 iterations while hybrid L2HC will do 20 method in each cluster. It must be ensured that the process is
iterations with 5 starting point for each iteration. This process run only once on each cluster, this is checked by the
is done 10 times in this work so the total iteration obtained is function. Can be concluded here that the
1000 iterations. other parameters used in L2HC model are ,
, and .
The challenging work in developing the proposed model is
to merge the part of clustering algorithm with the part of HC
searching. At this point, the searching part was inserted in the
(1) clustering looping fragment (see Code 3). It technically affects
that the proposed model operates two types of looping block
Preliminary Construct Data Experiment
for both checking the clusters and searching the optimal value.
Evaluation
Study Model Preparation (Testing)
Practically, it was dissimilar with [25] and [26] for catching
consensus rate, here we constructed a hybrid model via
Fig. 2. Experiment Stages.
combining two types of methods for obtaining new
optimization value for heuristic optimization problem.
After running 10 times process, the number of occurrence
of HC method to get the global optimum value was 73 and
hybrid L2HC got 85 times global optimum value from 1,000
iterations. From the obtained global optimum, 53.8% is
produced by the hybrid method, outperforming HC method
around 7.6%. Details of number occurrence for each process
can be seen in Fig 5.
The average of local optimum obtained from each iteration
is calculated and presented in Fig 6. We can see 8 out of 10
processes show hybrid L2HC method is better than HC. Also,
the average obtained by hybrid L2HC method is much more
stable. In other hands, HC produces an up and down average
value. This can be happened because LHC will divide
Fig. 3. Dataset Operated in Experiment. population into several clusters and randomize starting point
423 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
for each cluster. So the starting point for each iteration in the
process more or less will be in the similar position. Therefore,
it will lead to a similar local optimum value.
Code 3. Pseudocode of the Hybrid L2HC Method
Procedure LHS_HC()
Begin
initialize iteration
initialize numCluster
while(it<iteration)
<...variables definition...>
<...parameters cluster making...>
While(loopCount<numCluster)
eqStatus <-- true Fig. 6. Average Comparison of Optimum Value.
While(eqStatus = true)
cPos <-- random() TABLE. I. AVERAGE OF RUNNING TIME FOR EACH PROCESS
eqStatus <-- checkCluster(cPos)
End While
Hill Climbing Hyper-Cube-Hill-Climbing
search <-- true
cVal <-- objFunction(cPos) Running Time 0,001718695 0,002031107
While(search is not terminated)
bestNeighbor <-- 0,001718657 0,001718674
getBestNeigh(cPos,
totalData) 0,002031164 0,002968359
If(cVal>=bestNeighbor) 0,002031169 0,002031209
opt = cVal
//Local optimum is found 0,001562223 0,002187619
Search is terminated
Else //move to next position 0,001562517 0,002187636
cVal <-- bestNeighbor
0,002187026 0,001874907
cPos <-- bestNPos
End if 0,00203126 0,002030938
End While
loopCount++ 0,001718521 0,001718628
End while
it++ 0,002031105 0,001562419
End while Average 0,001859234 0,002031150
End
424 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
ACKNOWLEDGMENT [14] A. Cawsey, The essence of artificial intelligence. Prentice Hall PTR,
1997.
We would like to thank BINUS University who has [15] K. A. Sullivan and S. H. Jacobson, “A convergence analysis of
supported our works, particularly BINUS graduate program, generalized hill climbing algorithms,” IEEE Trans. Automat. Contr., vol.
Master of Computer Science. 46, no. 8, pp. 1288–1293, 2001.
[16] J. Boyan and A. W. Moore, “Learning evaluation functions to improve
REFERENCES optimization by local search,” J. Mach. Learn. Res., vol. 1, no. Nov, pp.
[1] P. R. Norvig and S. A. Intelligence, A modern approach. Prentice Hall, 77–112, 2000.
2002. [17] M. D. McKay, R. J. Beckman, and W. J. Conover, “Comparison of three
[2] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and methods for selecting values of input variables in the analysis of output
E. Teller, “Equation of state calculations by fast computing machines,” from a computer code,” Technometrics, vol. 21, no. 2, pp. 239–245,
J. Chem. Phys., vol. 21, no. 6, pp. 1087–1092, 1953. 1979.
[3] J. H. Holland, “Genetic algorithms,” Sci. Am., vol. 267, no. 1, pp. 66– [18] M. Stein, “Large sample properties of simulations using latin hypercube
73, 1992. sampling,” Technometrics, vol. 29, no. 2, pp. 143–151, 1987.
[4] R. Eberhart and J. Kennedy, “A new optimizer using particle swarm [19] J. C. Helton and F. J. Davis, “Latin hypercube sampling and the
theory,” in MHS‟95. Proceedings of the Sixth International Symposium propagation of uncertainty in analyses of complex systems,” Reliab.
on Micro Machine and Human Science, 1995, pp. 39–43. Eng. Syst. Saf., vol. 81, no. 1, pp. 23–69, 2003.
[5] A. Colorni, M. Dorigo, and V. Maniezzo, “Distributed optimization by [20] A. M. Briones, D. L. Burrus, J. P. Sykes, B. A. Rankin, and A. W.
ant colonies,” in Proceedings of the first European conference on Caswell, “Automated Design Optimization of a Small-Scale High-Swirl
artificial life, 1992, vol. 142, pp. 134–142. Cavity-Stabilized Combustor,” J. Eng. Gas Turbines Power, vol. 140,
[6] J. Cheng and M. J. Druzdzel, “Latin hypercube sampling in Bayesian no. 12, p. 121509, 2018.
networks.,” in FLAIRS Conference, 2000, pp. 287–292. [21] C. Maschio and D. J. Schiozer, “A new methodology for history
[7] K. Bryan, “Quasi-Newton Methods,” Rose-Hulman Inst. Technol., 2004. matching combining iterative discrete Latin Hypercube with multi-start
simulated annealing,” J. Pet. Sci. Eng., vol. 169, pp. 560–577, 2018.
[8] L. Rueda and V. Vidyadharan, “A hill-climbing approach for automatic
gridding of cDNA microarray images,” IEEE/ACM Trans. Comput. [22] A. Xhafa, E. Spaho, D. Elmazi, and M. Takizawa, “A Study on
Biol. Bioinforma., vol. 3, no. 1, p. 72, 2006. Performance of Hill Climbing for Router Placement in Wireless Mesh
Networks,” in 2015 10th International Conference on Broadband and
[9] D. N. Utama, N. Ani, and M. M. Iqbal, “An optimal generic model for
Wireless Computing, Communication and Applications (BWCCA),
multi-parameters and big data optimizing: A laboratory experimental
2015, pp. 460–465.
study,” in Journal of Physics: Conference Series, 2018, vol. 978, no. 1,
p. 12045. [23] B. Xi, Z. Liu, M. Raghavachari, C. H. Xia, and L. Zhang, “A smart hill-
climbing algorithm for application server configuration,” in Proceedings
[10] R. Malhotra, N. Singh, and Y. Singh, “Genetic algorithms: Concepts,
of the 13th international conference on World Wide Web, 2004, pp.
design for optimization of process controllers,” Comput. Inf. Sci., vol. 4,
287–296.
no. 2, p. 39, 2011.
[24] L. Hernando, A. Mendiburu, and J. A. Lozano, “Hill-Climbing
[11] H. E. Romeijn and R. L. Smith, “Simulated annealing and adaptive
Algorithm: Let‟s Go for a Walk Before Finding the Optimum,” in 2018
search in global optimization,” Probab. Eng. Informational Sci., vol. 8,
IEEE Congress on Evolutionary Computation (CEC), 2018, pp. 1–7.
no. 4, pp. 571–590, 1994.
[25] Y. Shang, "Hybrid consensus for averager-copier-voter networks with
[12] D. N. Utama, “The Optimization of the 3-d Structure of Plants, Using
non-rational agents," Chaos, Solitons & Fractals, vol. 110, pp. 244-251,
Functional-Structural Plant Models. Case Study of Rice(Oryza sativa L.)
2018.
in Indonesia,” Georg-August University, 2015.
[26] Y. Shang, "Consensus of hybrid multi-agent systems with malicious
[13] S. S. Rao, Engineering optimization: theory and practice. John Wiley &
nodes," IEEE Transactions on Circuits and Systems II: Express Briefs,
Sons, 2009.
pp. 1-1, 2019.
425 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—This study analyses different techniques used for research community recommends considering different
evaluation of various usability dimensions of software requirements when evaluating those applications, such as
applications (apps) being used on the smartphones. The scope of quantitative data (metrics), subjective evaluation (users'
this study is to evaluate various aspects of the usability impressions) and context data (e.g. environment and devices’
techniques employed in the domain of smartphone apps. conditions.).
Usability assessment methodologies are evaluated for different
types of applications running on different operating systems like Usability is known as a significant quality dimension to
Android, Blackberry and iOS, etc. Usability evaluation evaluate the quality and usefulness of smartphone applications
techniques and methodologies with respect to usability heuristics [2]. Usability is a quality attribute that assesses how easy and
like field experiments, laboratory experiments models and simple user interfaces are to use. According to Nielson,
usability standards are discussed in detail. The issues for usability can be defined as a method for improving the design
evaluation of usability of smartphone apps are identified by process. Usability is assessed on the basis of six dimensions
considering limitations and areas of improvement outlined in the which include learnability, memorability, efficiency,
contemporary literature. A conceptual framework for usability effectiveness, error rate and user satisfaction. Similarly, IEEE
evaluation of smartphone apps is also designed which would be Standard.610.12-1990 defines usability as “The ease with
validated through experimentation in the thesis work. This study which a user can learn to operate, prepares inputs for, and
is particularly useful to comprehend usability issues and their
interprets outputs of a system or component.” Usability
likely remedies to produce high quality smartphone apps.
dimensions affect four contextual factors which are users,
Keywords—Usability; Jakob-Nielson usability heuristics; technology, activity and environment. There are two
smartphone applications; ease of use; understandability; learning classifications of usability each having different number of
curve parameters. First classification includes effectiveness,
efficiency and user satisfaction which are part of ISO 9241-11
I. INTRODUCTION standard. Second classification includes understandability,
All Smartphones have become a daily use items and are learnability, portability and attractiveness, and is known as ISO
popular among all the sections of society. Smartphones are 9126-1 [3].
now the most popular mobile technology. The statistical data Table I illustrates six usability dimensions and their
show that one in each three citizens own a smartphones [1]. corresponding attributes. Brief description of these usability
Smartphones are in fact modern mobile phones with an classification attributes are:
additional highly developed computing capability and
connectivity. Their extensive modes for input are provided by a Effectiveness: How accurately the user achieves the
touch sensitive display. The most popular smartphones goals by using the app?
operating systems are Android, iOS, Windows phones and Efficiency: How much resources are consumed to
Blackberry. The quick and growing amount of smartphone perform certain tasks?
apps on the Google play and Apple stores have facilitated and
impelled the software experts to design applications of better User Satisfaction: How do the users sense about their
quality in order to compete in the markets. There are numerous experience by using the applications?
measureable aspects on the quality of software product and
usability is one of the most significant aspects [1]. Learnability: Is the system easy to learn? Novice users
should be able to complete basic tasks in a short period
In the recent years, the introduction of numerous of time with minimum training.
technologies has revolutionized our mode of communication,
entertainment and completing daily routine tasks. Memorability: The system is easy to
Simultaneously, the method of digital convergence has resulted remember/memorize. Users can return to it after a long
in the inventions of several devices like PDAs, smartphones, period of time and complete tasks without retraining.
tablets etc. which are able to gather different forms of human- Error Rate: If a user faces an errors while using the
computer interaction (HCI) in an integrated way. The HCI system, the system is capable to auto recovery.
426 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. USABILITY DIMENSIONS AND THEIR ATTRIBUTES 5) Error prevention (display appropriate mock up
Components of Nileson Usability Framework message).
6) Minimize the users memory load (interface should show
Usability Contextual Threshold values Classification
Dimensions Factors for dimensions standards the visible objects, actions and options).
1. ISO 9241-11 7) Efficiency of use and performance (animated icons and
Learnability transmissions should be displayed efficiently).
Effectiveness
Memorability
Users Time Efficiency 8) Aesthetic and minimalist design (avoid unwanted
Effectiveness
Technology Error rate User Satisfaction
Efficiency
Activity Number of apps 2. ISO 9126-1
information, particularly the one that is out of context).
User 9) Help and documentation (easy to find different content
Environment Number of users Understandability
Satisfaction
Error Rate
Learnability and documentation help is available).
Memorability 10) Help users identify, analyze and improve from errors
There are numerous ways to access the usability of (error should be displayed in user familiar language instead of
smartphone apps. The prominent methods include expert system language). Customization and shortcuts (shortcuts are
reviews, user testing, field experiment, laboratory experiments, available for frequently used actions).
system usability scale measurement and user surveys. The
Physical interaction and ergonomics (should provide
questionnaire consisted of four parts, i.e., demographic (This
physical sense like buttons, position, etc.) [6].
part of the questionnaire gathered the demographic facts of
participants, which includes age, gender, schooling, and If usability heuristics evaluations are conducted on a
earnings, and requested contributors to document their level of functioning/working product, the experts need to have some
enjoy with smartphones and methods to connect to the mobile specific user tasks in mind to focus the inspection in the right
Internet. Participants’ revel in with smartphones changed into direction [7].
measured with the aid of utilization hours in step with day and
the total utilization duration) statistics, customers’ preferences Usability is an important factor as if an app is difficult to
for the layout capabilities of smartphones, customers’ use then users would quit using that specific app. Usability
reputation of smartphones, and customers’ utilization behavior testing in a right manner, at the right time and with the right
of smartphones [4]. observation would reduce the software risks of building the
wrong product. For evaluation of smartphone applications,
Applying a heuristic evaluation approach using SMASH usability assessment is crucial so that developers can learn how
turned into shown to be effective in figuring out a huge to adopt them and consider the dynamicity of mobile scenarios.
percentage of the usability issues the aged customers
confronted at the same time as interacting with a smartphone. Usability evaluation of smartphone applications is a
The usability troubles had been not most effective because of potential research context that faces a number of challenges.
UI design, a number of the problems had been due to problems These challenges emerge due to the unique restricted features
of the elderly in performing the gestures which carried out to of mobile phone such as limited bandwidth, varying
the corresponding undertaking; particularly, the “drag and environmental factors and unreliable network [4]. Additional
drop”, and “faucet and keep” gestures. Therefore, it is challenges includes lack of usage-based testing and response,
recommended that the usage of those gestures be removed or at limited focus on interface architecture, navigation ignorance
least reduced. and connectivity, restricted resources and web connectivity
issues. The technical capabilities of mobile apps and achieving
Usability and user experience is a vital quality attribute for high level user satisfaction are crucial for the success of mobile
apps. The hedonic aspects such as fun, emotions and apps. Hence, usability testing of mobile apps is mandatory
enjoyment are focused by user experience. Hedonic aspects process to ensure that mobile apps are practical, effective and
meet up the universal needs, but they do not essentially have a easy to use [8].
utilization worth. User-ability is another aspect which is
integrated with the user experience to determine whether the Because of the significance, a large number of usability
user felt pleasant or otherwise with the system during the guiding principles have been modeled to support the structure
inspection. An approach is used by the users in which they of usable applications. These guiding principles are specially
draw and write their hedonic aspects about the application [5]. proposed for web-based and desktop applications. Due to the
This will leads to map the usability attributes on the usability mobility nature of smartphones devices, the smartphone apps
heuristics as follows: are different in many ways from the conventional applications
[9]. To date, the guiding principles for usable smartphone
1) Visibility of the system (system should keep the user applications are isolated and limited. This adds difficulty and
informed about all processes and state changes through complexity to evaluate the usability of smartphone apps [10].
feedback within a reasonable amount of time). This aim of this study is to explore the various usability
2) Match between system and real world (should speak dimensions and the corresponding issues that need to be
users language instead of system oriented language). considered while designing and evaluating mobile apps. In this
3) User control and freedom (Allow users to undo and study, we evaluate different usability dimensions along with
redo the previous tasks). their testing parameters that are necessary to be adhered to
4) Consistency and standards (the user should be able to ensure better quality of the mobile apps and their user-
do things in a familiar standard and consistent way). friendliness.
427 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
II. LITERATURE REVIEW usability using a randomly selected model from other mobile
Moumane et al. [3] present empirical evaluation for the use user population based on different types of branded apps like
of software quality standards ISO-9126 in mobile native, hybrid and desktop apps.
environments and highlight issues related to the software such Lorusso et al. [15] address the usability and learnability of
as user guides, use of simple data entry procedure and NFC based application. This research aims to explore the user
existence of online help. The study performs hardware based satisfaction, usability, learnability and quality of the interaction
evaluation such as display resolution, memory size and screen between the children who have language disorder and the
size by using ISO-9241 and ISO-5062 standards benchmark. system application. Autonomy level, feedback, satisfaction and
The authors analyze two usability evaluation standards for learnability are usability variables. Limited numbers of
mobile apps including ISO-9241and ISO-25062.ISO standard activities are offered by the selected prototype/hypothesis. As a
9241is a base quality model and includes efficiency and future dimension, a systematic research should be conducted in
satisfaction as the usability dimensions. The level of user educational environment.
satisfaction against three usability factors were evaluated as
62%,33% and 20% respectively. ISO standard 25062 includes Hussain et al. [16] explore the usability on kindle app for
two usability factors reliability and portability. The proposed smartphone platform. For this purpose usability attributes as
framework was compared with ISO standard 9126 on the basis visibility, efficiency, satisfaction and enjoyability. Study
of three usability measures efficiency, effectiveness and proposed a descriptive based statistical methodology to
satisfaction by conducting a survey. evaluate the usability attributes. In laboratory based experiment
15 participants were chosen randomly with different age
Sorber and Kortum [11] addressed the subjective usability groups ranges as 18-29, 30-39 and 40-49. Five tasks were
of a large number of mobile apps for both tablets and phones performed by the participants. One minute is set for execution
across Android and Mac operating systems and target the of task. The test session is recorded by video camera. The front
consistency measure in this context. The objective of the study screen recorded the emotions, time, error and navigation from
is to propose a baseline for usability measures for mobile apps. one page to another page of the users. A post-test questioner is
The author describe the usability on a large number of mobile given to the users. There were three measurements such as
apps for both tablets and phones across Android and Mac time, error, and frequency. Quantitative data is gathered by the
operating systems; characterize these results on system test results and analyzed with the SPSS. Descriptive statistics
usability scale and describe the usability measures for include Min, Max and Standard deviation were used to analyze
consistency. The proposed solution is used only on small scale. and present the resulted data, these resulted values are mapped
Future work could be examining the usability of mobile apps in on Likert scale which has values as strongly agree, Disagree,
a more formal laboratory based environment. Natural, Agree and Strongly disagree. Future studies could be
Lu and Wei [12] analyze the level of enjoyment, mobility conducted on large no of participants when the sample is
use towards the persistence use of smartphone apps. The goal projected to the large population.
of study is to revise the IS (Information System) continuance Nascimento et al. [5] explore the usability by addressed the
model to highlight the role of enjoyment and mobility of user relationship between user experience and usability. Study
perception towards the continuance of user intention to use the proposed a technique “userability” to evaluate the usability for
smartphone application. The IS continuance model is based on mobile applications. The userability is the integration of user
the ECT(Expectation Theory of Continuance).ECT is mostly experiences and usability which helps to the developer
used in the literature for consumer behavior to analyze the designer as well as non specialist in the domain of human
consumer satisfaction, service marketing in general and post- computer interaction. Proposed methodology consists of two
purchase behavior(e.g., repurchase, complaining), (Anderson further steps as heuristic evaluation for usability and 3E
and Sullivan1993;The usability measure include satisfaction, method. The 3E method stands for Expressing Emotions and
performance expectancy, continuance intention, post usage Experiences. Study define the ten aspects from heuristic
attitude and effort expectancy. The study defines the context of evaluation and from 3E method the two questions are finalize
smartphone apps broadly and data were analyzed on a smaller for evaluation as "what users feel regarding ten heuristics
scale. Future work could be to validate the proposed model on aspects and "what are the improvements did users feel for this
large scale by using some other design methodology. aspects. Satisfaction is the main attributes in 3E and ten
Baek and Yoo [14] evaluate the usability attributes for heuristic aspects for this evaluation emojis of face expression is
"branded mobile applications" to measure and conceptualize used. The satisfaction attribute is scaled on questioner as
the underlying scope of usability. The research objective is to unsatisfied, moderately satisfied, little satisfied very satisfied
propose and examine the measurement tool to explore the and very satisfied. The proposed methodology phases include
perceived level of usability application. Study proposed the the steps as Training, Application scope, scope of the activities,
usability factors as user friendliness, omnipresence speed, fun qualitative analysis of inspection questioner Detection issues
and personalization, for app and proposed a valid and reliable and Data analysis. Five applications are used measurement are
exploration of the usability application that included user time, no of errors, no of duplicated issues, suggestions and
perceptions. The study develops the usability scale duplicated suggestions. Grounded Theory method is used for
development model. Variable of interests in data gathering validation to perform data coding.
were explored and measured using self-report survey. Future Salman et al. [6] evaluate the usability of user interface for
study could be conducted to validate the usability scale for smartphone applications used by elderly. Heuristic evaluation
428 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
methodology is used for this purpose. Issues are highlighted standard deviation. These studies result is generating by
during the expert testing sessions. Identified problems are considering the android operating system on apple or any other
classified into different categories as appearances, language, operating system results may differ or better just because of
dialogue and information. These classifications are further hardware limitation as display resolution speed, etc.
divided into sub categories and proposed solution for user
interfaces. From heuristics evaluation two heuristics are III. CRITICAL ANALYSIS
frequently violated as minimize the user memory load and In this study after reviewing the reviewing the literature
match between system and real world. The test session is the usability assessment methods for testing are briefly
performed by the 5 experts who have different age groups and described and critically evaluate the literature to explore
have different qualification in human interaction domain on various usability dimensions as well as corresponding
smartphone model J7. By the end of the results, 27 problems challenges that need to be considered while designing and
are highlighted extracted from a checklist. In future studies, a evaluating the mobile applications. By considering the various
think aloud session would be conducted with the elderly usability dimensions, attributes, performance measures,
participants at the time of development and design a prototype contextual factors, testing parameters, proposed model and
after getting their relevance feedback. validation model as well.
Lee and Lee [13] evaluate the usability attributes for The attribute of the critical table are Problem addressed,
augmented reality mobile application. This research aims to Usability dimension, Implementation method, platform used
develop a tool for creating user-based design interfaces in for application, standard model for comparison or validation
mobile augmented reality (MAR) education and check list for purpose, mapped with Nielsen findings for user interface
usability with factor analysis and reliability. Examine the named as SMASH (smartphone usability heuristics) and
usability attributes of multimedia AR and to develop a usability limitation or future findings.
evaluation tool through concretization. This study examined
the usability attributes like learnability, ease of memory, usage IV. KEY CHALLENGES
convenience and satisfaction. The evaluation items collected Since mobile technologies are used in every field,
from existing research were used as basic data for developing smartphone apps play a vital role for their success. So usability
the usability evaluation checklist survey was conducted with is a crucial factor to achieve the quality goal but usability
122 experts, and after factor analysis and reliability analysis, testing of smartphone apps faces number of challenges like:
the final checklist for each usability evaluation item was
prepared. Affordance and presence are main measures of Connectivity [3,8]: The slow, unchangeable network
reliability with cognitive affordance, sensible affordance, and association with small size bandwidth is an ordinary difficulty
physical affordance .proposed the usability evolution tool with for smooth execution of smartphone apps. The difficulty
focus group interview, factor analysis and reliability on mainly affects loading time and worth of stream media.
evolution question. Proposed evolution steps do not validate Small screen size [3,8]: The diversity of element ratio and
with the existing model. Future work could be to validate the pixel solidity can be massive.
proposed usability steps on large no of applications in formal
laboratory experiment. Different display resolutions [4]: As different display
resolution may produce different usability evaluation results,
Liu et al. [17] address the usability aspects under short resolution can disgrace the quality of information display
acceptance and usage behavior of smartphone applications. on mobile devices.
Factors are analyzed regarding acceptance and behavior.
Questioner was developed which is filled by 842 participants. Context of mobile [14]: It may illustrate as any statistics
Acceptance is measured as usefulness, ease of use, and that differentiates a situation linked to conversation between
intention of use. Nine factors are explored for acceptance as user, apps and the encompassing historical past. It naturally
element for design interface, physical smartphone consists of location identities of close by humans, gadgets and
characteristics, feedback for touch screen, display screen, environmental basics that could divert person awareness.
connectivity and application. A questioner is constructed
Capability and Limited processing power [15]: Some
comprising four parts demographic information (gender, age,
smartphone apps want large quantity of remembrance to GUI
education and income etc.), user preferences, features of
assist which include three dimensional apps.
smartphone (icon size, icon color, shape, font size, character
spacing, etc.), users acceptance and their usage behavior (task Lesser focus on navigation [14]: Interface structure play a
based like shopping Skype chatting, etc.). Data were collected critical position to get the consumer pride stage. On interface
an online survey of smartphone users from a Chinese website. structure there's a number of unnecessarily links or buttons
For data analysis EFA (exploratory factor analysis) is used to which burdened the consumer and frustrated pop-up messages
detect the critical design features factor and CFA (confirmatory on each second.
factor analysis was) is used to check reliability and validity for
measurement constructs.842 participants including 378 male Lack of use testing and response [8,15]: The most
and464 are female having at least 4 years’ experience for using important challenges in usability testing is the lack of user
smartphone, age groups ranges as 20 to 51 years. Age attribute testing and their feedback during the design evaluation process.
is divided into categories then find frequency of each age There is a need to get the acknowledgement from user to
groups among male and female and result the mean and determine their needs, intentions, usability obstacles through a
descriptive or statistical measure.
429 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Not strengthening the engagement loop [3,15]: Designer Selection usability attributes: In the first step usability
should carefully design the app by getting the user experiences, selection is the initiation process to usability of smartphone
their preferences for the application by performing the tasks app. The user selects the usability dimensions as per
and taking the feedback. requirement or his scope of the work. The usability attribute
are stored as a list in the tabular form as learnability efficiency
V. CONCEPTUAL FRAMEWORK memorability, error, satisfaction and effectiveness,
Conceptual framework for usability evaluation of effectiveness, efficiency and user satisfaction are mostly and
smartphone is described in Fig. 1. commonly used as per literature review. so selection the
usability is a vital for further processing.
Conceptual framework is the approach to represent a
general concept, that guide people appreciate or reproduce the Usability evaluation based criteria: In this phase the
domain of that model. Conceptual models illustrate the usability evaluation based criteria is chosen to access which
relations between factors and the stream of data or processes. A standards or models are the based for further processing. The
conceptual model comprises the four fundamental criteria list for usability testing as standards and model like
characteristics as the potential reward to implement a (ISO 9241-11 and ISO 9126-1) are stored in the tabular form
theoretical model are numerous, but mostly depend on the own with their different usability parameters.
capability to invent a well-built model in the primary place. Fetch threshold values for usability attributes: In the third
The key rewards of theoretical model include. step after selection of usability dimension and criteria the
Description: The above conceptual model is designed for threshold values are selected. The threshold values must be
usability testing of Smartphone applications. The conceptual defined to meet or to compare the mature results like volume of
model comprises of six steps as selection of usability attribute, expected traffic, error rate, time etc. The threshold values may
usability evaluation based criteria, fetch threshold values for be quantitative or statistical measure like time, frequency, min,
usability attributes, select the application for usability testing, max, error rate, number of users and number of apps, etc. The
design test case generation on the basis of usability criteria, threshold measurement values are fetched from a table in
Test execution and check the criteria meet for validation. Three which these values are listed.
mini table which support to select the attributes as (learnability, App selection: After set the usability dimension their
efficiency, memorability, error, satisfaction and effectiveness) testing parameters there is need to implement on an
standards and models to follow for the basic initiation process application. The application is selected on bases of the nature
as (ISO 9241-11 and ISO 9126-1) and measuring parameters as of the testing. The contextual factors as technology, type of
(time, error rate, number of users and number of applications users, activity and environment etc. play a vital role for
laboratory or field experiment, and contextual factor as well) selecting the application. Which platform, operating system is
for usability testing as stored in a tabular form. All these steps used for testing this will be set in this phase.
should be followed before designing the smartphone
application. Arrows show the flow of data, boxes indicate the Test case generation on the base of usability criteria: In the
main steps, diamond symbol show the decision in form of 4th step when the application, platform operating system is
“Yes” or “No”. selected then there’s a need to design a prototype for testing.
Prototype is usually used to estimate a brand new design to
growth accuracy with the aid of device examination and
consumer. A prototype is an premature pattern, reproduction or
freed from a product built to check a belief or procedure or to
carry out as a thing to be simulated or found out from. This can
be questioner, laboratory, field experiment or in a controlled
environment.
Test Execute: After selecting the platform, the operating
system and the application the designed prototype is being
executed by the users. The method of execution could be a pre
or post questioner, advance techniques like eye tracing and
facial recognition is used to get user experiences, and emotions
“3E method” is used which explore the expressing emotions
and experiences.
Usability criteria met: In the last step execution of the
designed prototype and calculate the result after the processing.
These results are validated with an existing model or prototype
result. Results are noted and validate with the models that is
chosen on the second step as criteria for usability evaluation. If
Fig. 1. Conceptual Framework for usability Evaluation. the criteria meet the with the model or show better
430 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
performance and then results are documented in the form of [3] K. Moumane, A. Idri and A. Abran, "Usability evaluation of mobile
report But if the criteria is not meet then move back to the step applications using ISO 9241 and ISO 25062," Springerplus, 5(1), 2016.
one and repeat the process by changing the attributes or [4] H. Rahmat, H. Zulzalil and A. Abdulghani, "An approach towards
development of evaluation framework for usability of smartphone
performance parameters for good results. evaluation results are applications," Malaysian Software Engineering Conference (MySEC),
met with the defined criteria or enhancement is occur while pp. 178-182, 2017.
comparing then report is generated ,if the evaluation results are [5] I. Nascimento,W.Silva, B. Gadelha and T.Conte, "Userbility: a
not met with the pre-defined criteria. technique for the evaluation of user experience and usability on mobile
applications," Human-Computer Interaction, 37:3, 372-383,19 June
VI. CONCLUSION AND FUTURE WORK 2016.
[6] H.M. Salman, W.F. Wan and S. Sulaiman, "Usability evaluation of the
Usability is recognized as a significant quality dimension to smartphone user interface in supporting elderly users from experts’
determine the success of mobile applications. This study perspective," IEEE Access, 6, pp. 22578-22591, 2018.
highlights the techniques which are being applied to evaluate [7] D. Quiñones and C. Rusu, "How to develop usability heuristics: A
the usability of smart phone applications. Usability assessment systematic literature review," Computer Standards & Interfaces,
methodologies are evaluated for different types of applications 53(2017): 89-122, 2017.
running on different operating systems like Android, [8] B.A. Kumar and P. Mohite, "Usability of mobile learning applications: a
Blackberry and iOS etc. Assessment methods of usability systematic literature review," Journal of Computers in Education, 5(1),
testing are discussed in a great detail to explore various pp. 11-17, 2018.
usability dimensions as well as corresponding challenges that [9] R.Alturki, V.Gay and R. Alturki, "Usability attributes for mobile
application: a systematic review," 7th International Conference on
need to be considered while designing and evaluating the Computer Science, Engineering & Applications, 2017.
mobile applications. Specifically, the study conducted a critical [10] D. Zhang and B. Adipat "Challenges, methodologies, and issues in the
review of various usability dimensions, attributes, performance usability testing of mobile applications," International Journal of
measures, contextual factors, testing parameters, proposed Human-Computer Interaction, 18(3), pp. 293-308, 2005.
model and validation model as well. The prominent challenges [11] P. Kortum and M. Sorber, "Measuring the usability of mobile
identified in this study include: connectivity, small screen size, applications for phones and tablets," International Journal of Human-
different display resolution, information input method, context Computer Interaction 31(8), 518-529, 2015.
of mobile, capability and limited processing power, navigation [12] J. Lu, C. Liu and J. Wei, "How important are enjoyment and mobility
ignorance, no focus on interface architecture etc. A conceptual for mobile applications?," Journal of Computer Information Systems, 57
(1), pp. 1-12, 2017.
framework for usability evaluation of smartphone apps is also
[13] W-H Lee and H-K Lee, "The usability attributes and evaluation
designed which would be validated through experimentation in measurements of mobile media AR (augmented reality)," Cogent Arts &
the thesis work. This study is particularly useful to comprehend Humanities, 3(1), 2016.
usability issues and their likely remedies to produce high [14] T.H. Baek and C.Y. Yoo, "Branded app usability: conceptualization,
quality smartphone apps. The study provides a conceptual measurement, and prediction of consumer loyalty," Journal of
framework for usability testing of smartphone applications. Advertising, 47(1), pp. 70-82, 2018.
Future work could be conducted to validate this model in a [15] M.L. Lorusso, E. Biffi, M. Molteni and G. Reni, "Exploring the
formal manner. learnability and usability of a near field communication-based
application for semantic enrichment in children with language
REFERENCES disorders," Assistive Technology, 30(1), pp. 39-50, 2018.
[1] B.C. Zapata, J.L. Fernández-Alemán, A. Idri and A. Toval, "Empirical [16] A. Hussain, E.O. Mkpojiogu, J.A. Musa and S. Mortada, "A user
studies on usability of mhealth apps: a systematic literature review," experience evaluation of Amazon kindle mobile application," In AIP
Journal of medical systems, 39(2), 2015. Conference Proceedings (Vol. 1891, No. 1, p. 020060), AIP Publishing,
[2] R. Inostroza, C. Rusu, S. Roncagliolo, V. Rusu and C. Collazos, 2017.
"Developing SMASH: a set of smartphone'susability heuristics," [17] N. Liu and R. Yu, "Identifying design feature factors critical to
Computer Standards & Interfaces, 43 pp.40-52, 2016. acceptance and usage behaviour of smartphones," Computers in Human
Behaviour, 70(2017), pp.131-142, 2017.
431 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In globalization of information, internet has played remembering the passwords, tend to use either minimum
a vital role by providing an easy and fast access of information length allowed for passwords or use common words, names or
and systems to remote users. However, with ease for authentic simply write them down in a notebook or a system file or use
users, it has made information resources accessible to same password for multiple personal accounts, which makes
unauthorized users too. To authorize legitimate user for the the attacks like, dictionary attack, brute-force attack, hybrid
access of information and systems, authentication mechanisms attacks, social engineering attack, dumpster diving attack,
are applied. Many users use their credentials or private shoulder surfing and key logging attacks possible [17, 19].
information at public places to access their accounts that are
protected by passwords. These passwords are usually text-based To overcome the limitations of Textual passwords,
passwords and their security and effectiveness can be Graphical passwords are developed and used as an alternative
compromised. An attacker can steal text-based passwords using method for authentication purpose [3]. As the name implies,
different techniques like shoulder surfing and various key logger this authentication method makes use of sequence of images
software, that are freely available over internet. To improve the or shapes instead of text, as the password.
security, numerous sophisticated and secure authentication
systems have been proposed that employ various biometric Graphical password overcomes the drawbacks of textual
authentication systems, token-based authentication system etc. password. Studies have shown that human brain can retain
But these solutions providing such high-level security, require images more easily as compared to text [4, 5], and this
special modification in the design and hence, imply additional property entitles graphical passwords as a more easily
cost. Textual passwords that are easy to use but vulnerable to memorable method [6]. It is comparatively secure than textual
attacks like shoulder surfing, various image based, and textual password against dictionary, brute-force, social engineering
graphical password schemes are proposed. However, none of the and key-logging attacks [2] but vulnerable to shoulder-surfing
existing textual graphical passwords are resistant to shoulder attack [6, 20] where authentic user is observed while entering
surfing and more importantly to mobile key-logging. In this the password [7].
paper, an improved and robust textual graphical password
scheme is proposed that uses sectors and colors and introducing Graphical password systems can be classified as either
randomization as the primary function for the character display recognition-based or recall-based approach; the latter of which
and selection. This property makes the proposed scheme is further divided into cued recall-based and pure recall-based
resistant to shoulder surfing and more importantly to mobile approach [2, 7, 18]. In the recognition-based approach, the
key-logging. It can be useful for authentication process of any user selects a set of images from the available images in the
smart held device application. registration phase which are recognized and reselected in the
same sequence in the login phase. In the second approach, i.e.,
Keywords—Authentication; graphical password; shoulder recall-based, the user recalls something that was selected in
surfing; mobile key-logging; security the registration phase. For this process there might be a clue
I. INTRODUCTION given to the user–cued recall-based or no clue given at all
during login phase–pure recall-based approach; the former of
Access control mechanisms are widely used to protect user which is easy to use.
resources especially information asset. The legitimate user is
required to authorize himself by passing the authentication The existing graphical password schemes are divided into
technique employed on the system [1]. The conventional and two types; image-based graphical passwords [2, 7, 8] and
widely used authentication method is login system protected textual graphical passwords [1, 7]. In image-based, as name
with a textual password [2]. It is a variable length combination implies, images / symbols are used for the password. Whereas,
of alphabets, digits and special characters. Though it provides the textual graphical password consists of a pie shape
considerable security level, this approach has its containing colors and sectors, which further contains different
shortcomings. To make a textual password robust against characters for selection as password. However, image-based
various password-based attacks, user has to select random graphical passwords are susceptible to shoulder-surfing attack
characters, and some authentication systems require the user as images can be retained easily in mind [4]. Many textual
to change the password frequently. Users for their ease in graphical password schemes are developed but they lack
432 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
efficiency in terms of login time and robustness against L. Y. Por et al [13] presents a password scheme that is
shoulder-surfing and mobile key-logging attacks. based on „digraph substitution rules‟ that hides the activity
performed to drive the password images. This scheme is
In this paper, we have proposed a textual graphical resistant against shoulder surfing attacks as user clicks only on
password scheme for smart held devices that is resistant to one of the pass images instead of both pass-images. However,
shoulder surfing and mobile-key logging attacks. This scheme this scheme requires the user to know the digraph substitution
is a combination of recognition-based and pure-recall based rules hence not very user friendly. Another graphical
approach and incorporates randomization on every text password scheme is presented in [14] by GC. Yang in 2017,
character with a click. The rest of the paper is organized as and its improvement in 2018. This scheme is based on pass-
follows. Section II reviews the related work, Section III position scheme which is similar to pass-point approach
explains the working of existing system. Section IV presents however in pass-position a relative value of the clicked
our proposed system and scheme. Section V gives an analysis location is also accepted rather than an exact value. This
of the proposed scheme and lastly, Section VI concludes the feature makes this approach user friendly but on the other
paper. hand prone to accidental logins, hence less secure.
II. RELATED WORK A. Mishra et al. [6] have presented an image-based
S. Wiedenbeck et al. [9] proposed „The Convex Hull Click graphical password scheme which is based on falsification
(CHC)’ scheme which is a game like graphical authentication method. This scheme is resistant to graphical password attack
method where user without clicking on the images can select but has poor security against key-logging attacks as user has to
the graphical password in an unsecure environment. However, enter the credentials using a keyboard. Another image-based
this scheme entails a longer authentication process. The login graphical password scheme is presented by K. Irfan et al. [15]
time consumption is reduced in a scheme presented by H. Gao which use both image-based and test-based approaches. The
et al. [10] where the author has proposed a graphical password user selects few images on registration which are reselected on
scheme based on ColorLogin. In this scheme a group of login. If these images match with the images stored in
chosen-color icons are displayed for the user to set as his pass- database, the user is asked to change his password by selecting
icons. The drawback of this scheme is a comparatively smaller new images. This approach makes this scheme less practical
password space and most importantly impracticality for colour for changing the password on every login.
blind users.
III. EXISTING SYSTEM
Prof Raut et al. [7] have presented another graphical The existing system [1] is a type of textual graphical
password scheme that is also based on colours. This proposed password scheme that consists of pie that is divided into 6
scheme combines colours with textual characters in a pie sectors each having 12 randomly distributed characters in
chart. The user selects a colour and a sector as his password, them. The divided sectors contain 72 characters in all; having
however, only the characters fixed in a sector can be selected 26 alphabets (26 upper case and 26 lowercase), 10 special
limiting the choice of characters for the user. A similar characters and 10 decimal digits (0-9). The login page of the
scheme is presented by Sumit H. et al. [1] which does not existing scheme is shown in Fig. 1. The user verification is
make use of colour in the pie chart. This scheme has the same done in two stages.
limitation of fixed set of characters in any sector for a user to
select. An image-based graphical password scheme is A. Registration Phase
proposed by Pooja K. S. et al [8], for ATM systems where The registration phase consists of the following steps;
user has to select a sequence of images out of 16 images. The
scheme provides shuffling of these images on every login but 1) The user selects one of the six sectors from the given
is prone to Hidden-camera attack. pie shape, that is used as the pass-sector for further logins.
Another image-based scheme is proposed by E. Darbanian 2) The selected sector has 12 of the 72 randomly placed
et al. [11] in which a set of images are selected which are characters that constitutes the textual password of the user.
interpreted as characters at the back end. Each displayed 3) The sector number and the textual password is
image holds a value of a character that is translated by a pre- encrypted and stored in the password table in the system.
defined table. This scheme, however, is complex as user has to
memorize the characters as well as the associated images.
Mrs. Aakansha S. et al. [2] have presented an image-based
graphical password scheme which is a combination of
recognition and recall-based approach. In this scheme user is
presented with a set of images and questions. The user has to
select a number of images and three questions that are
answered by clicking a specific point on the given images.
This scheme provides a large password space but is inefficient
in terms of time required for the login process. Another
graphical password scheme is proposed by A. Ahmad et al. in
[12] that comprises of textual characters shown in a grid. This
scheme is tested to be robust against shoulder surfing attack
but has high complexity and is not user-friendly. Fig. 1. Login Screen of the Existing Scheme [1].
433 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
1) Upon login the system displays the pie shape having 6 The registration phase consists of the following steps:
sectors. 1) On the registration page (Fig. 4(a)), user is asked to
2) The system places the 72 characters equally divided enter first name, last name and a valid email address. The user
among all the sectors. email will be used as username.
3) The user has two buttons “Clockwise” and “Anti- 2) After filling these fields, the user selects the ‘Generate
clockwise” available. password’ button and a new screen appears containing the pie
4) The user by using these buttons in desired direction chart with 8 sectors each having a different colour that are
rotates the sectors in order to match the selected sector number selected separately (Fig. 4(b)).
and the characters of the set password. 3) There are four buttons below the pie chart for the
5) When the desired sector coincides with the sector character selection:
number, the user clicks the “Confirm” button and is allowed to a) First button is of upper-case letters (A... Z), by
successfully login into the system. touching this button, upper case letters will randomly display
The existing system offers a fixed set of characters in each in the sectors of the pie chart.
sector of pie for the registration or login phase. Thus, making b) Second button is of numbers (0… 9), by touching this
the characters predictable, selected as password, as user has to button, numbers will randomly display in the sectors of the pie
select whole sector as the pass-sector. Secondly, a fixed sector chart.
containing the complete set of characters of password makes c) Third button is of special characters (#*$%), by
this approach vulnerable to mobile key-logging attack. touching this button, special characters will randomly display
IV. PROPOSED SCHEME in the sectors of the pie chart.
d) Forth button is of lower-case letters (a… z), by
The proposed scheme is an improvement over the existing
touching this button, lower case letters will randomly display
scheme and overcomes its shortcomings. It is a combination of
recognition-based and recall-based graphical password in the sectors of the pie chart.
scheme. Similar to the existing scheme, the proposed scheme 4) User selects the desired color and then selects
is also based on text and sectors contained in a pie. The characters for his password meeting the minimum requirement
architecture of the proposed scheme is shown in the Fig. 2. It set for the password. The colour and characters are selected by
has been developed for Android based smart-held devices simply touching the screen on the desired point.
using Android studio, languages used are Java and Android. 5) Upon submitting the selected graphical password, the
The user selects the password from a range of 72 user Email and password are hashed using SHA-1 and stored
characters randomly distributed over 8 sectors during in database.
registration and login phase. The character set contains 26 6) After the complete registration process, login page
upper-case and 26 lower-case alphabets, 0-9 digits and 10 appears.
special characters (@, !, #, $, %, *, &, ?, <, >). For securing Fill form,
enter email
the user credentials, the selected password along with other & set
graphical
user-entered information, are hashed using hashing algorithm password Registration Password Name,
hash Email &
SHA-1 [16] and stored in the database, during both hashed
password
registration and login phase. Moreover, characters are always Hashing
At login
shuffled at the run time which will ensure security against email & Database
User
shoulder surfing and mobile-key logging. Mobile Password
password
verify from
Enter Login
hash database
The proposed scheme comprises of a user registration and email &
password
login phase that are described as follows: Verified output
requirement for the password is 8 characters that must contain Fig. 3. User Registration.
434 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Enter Email
Select Pal(L)
password
Mobile
Verify Email
Password C. Robust against Shoulder Surfing and Mobile-Key logging
Will hash
User and hashed
password
Attack
Verified output SHA 1 Get password
Check mail
The proposed scheme is robust against shoulder surfing
Database and mobile-key logging attacks. The existing schemes have a
Fig. 5. User Login Process. fixed set of characters displayed in each sector after the first
435 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
display which shows them randomly but after that, they are The proposed scheme provides high security against
fixed for the whole session of registration or login whereas in Brute-force attack as it offers a very large password space as
our proposed scheme, the randomization works on every compared to the existing scheme. The randomization feature,
single click till the length of the password to be selected incorporated with every click, adds robustness against
during registration or login. shoulder surfing and mobile-key logging attacks. In case an
authentic user forgets his password, the password is email to
This randomization works independently on each character the user hence adding another layer of security and making an
and places them in a different sector upon every click thus attacker unable to get hold of the password. The proposed
decreasing the probability of a sector being selected in the authentication graphical password scheme is designed for
same sector with the same characters. This increases the smart held application and can be easily used as a secure
security level of our scheme as character position is totally gateway for any application.
unpredictable.
The work can further be extended by increasing the
Let probability of a character to be placed in any sector to number of characters and sectors in the pie, and by also
be denoted by P(c) which can be calculated as shown in (3) increasing the number of colours of the sectors. To further
and (4). enhance the robustness of this scheme, two factor
(3) authentications can also be incorporated.
REFERENCES
(4) [1] S. H. Wagh, A. G. Ambekar, “Shoulder Surfing Resistant Text-based
Graphical Password Scheme”, ICCT 2015, International Journal of
This factor plays an important role against Shoulder Computer Applications (0975 – 8887).
Surfing attack and mobile-key logging attack as the [2] Mrs. A. S. Gokhalea, Prof. V. S. Waghmare, “The Shoulder Surfing
Resistant Graphical Password Authentication Technique” 7th
probability for the attacker to correctly guess any character on International Conference on Communication, Computing and
a given location is substantially low even if he tried to Virtualization 2016, Procedia Computer Science 79 (2016) 490 – 498.
memorize the password characters and tries to login into the [3] X. Suo, Y. Zhu, G. Scott. Owen, “Graphical Passwords: A Survey”,
system. Department of Computer Science Georgia State University. 21st Annual
Computer Security Applications Conference (ACSAC‟05), IEEE
Another important factor that makes our scheme strong [4] Kirkpatrick. “An experimental study of memory”, Psychological
against shoulder surfing attack is the rotation feature of the Review, 1:602-609, 1894.
pie chart. The user can rotate colour rim and the sectors [5] R. Shepard. “Recognition memory for words, sentences and pictures”,
independently in clockwise or anti-clockwise direction. The Journal of Verbal Learning and Verbal Behavior, 6:156-163, 1967.
selected colour and the sectors don‟t have to be aligned to [6] A. Mishra, R. Jadhav, S. Patil, “A Shoulder-Surfing Resistant Graphical
enter the password. Password System”, International Research Journal of Engineering and
Technology (IRJET), Volume 5, March 2018.
This feature enhances the security of our scheme against [7] Prof Raut S.Y., J. B. Baviskar, K. Rahul S, S. Aditya N, S. Yogesh S,
shoulder surfing attack. An attacker cannot guess the “Shoulder Surfing and Keylogger Resistant using Graphical Password
associated colour with the characters as they need not be Scheme”, International Journal of Advanced Research in Computer
aligned with any particular sector for the password to work. Science, Volume 5, No. 8, Nov-Dec 2014.
[8] Pooja K S, P. V. Dhooli, Prathvi, Prof. Ashwini N, “Shoulder Surfing
The proposed scheme provides strong authentication Resistance Using Graphical Password Authentication in Atm Systems”,
process in case of password recovery scenario. If a user International Journal of Information Technology & Management
forgets his password, the system sends the password to the Information System (IJITMIS), Volume 6, Issue 1, January - June
(2015), pp.01-10.
linked email account of the user. This is very beneficial for the
authentic user but any imposter trying to login into the system [9] S. Wiedenbeck and J. Waters, L. Sobrado, J. C. Birget, “Design and
Evaluation of a Shoulder-Surfing Resistant Graphical Password
cannot access the system in any way. Scheme”, AVI '06, May 23-26, 2006.
Thus, our proposed graphical password scheme is highly [10] H. Gao, X. Liu, R. Dai, S. Wang and H. Liu, “Design and Analysis of a
secure and easy to use. Graphical Password Scheme”, Fourth International Conference on
Innovative Computing, Information and Control, 2009.
VI. CONCLUSION [11] E. Darbanian, Gh. D. Fard, “A Graphical Password Against Spyware
and Shoulder-surfing Attacks”, International Symposium on Computer
User authentication plays a vital role in securing user Science and Software Engineering, IEEE, 18-19 Aug. 2015.
accounts and confidential information. In this paper, a new [12] A. Ahmad, M. Asif, M. Hanif, R. Talib, “Secure Graphical Password
graphical password user authentication scheme for smart-held Techniques against Shoulder Surfing and Camera based Attacks”,
devices is presented which is a combination of recognition and International Journal of Computer Network and Information Security ·
November 2016.
pure recall-based graphical password approach. With a
[13] L. Y. Por, C. S. Ku, A. Islam, T. F. Ang, “Graphical password: prevent
combination of colour and alphanumeric characters, this shoulder-surfing attack using digraph substitution rules”, Higher
scheme is viable for users comfortable with textual passwords. Education Press and Springer-Verlag Berlin Heidelberg, 2017.
436 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[14] GC Yang, “PassPositions: A secure and user-friendly graphical [18] GC Yang, H. Oh, “Implementation of a Graphical Password
password scheme”, 4th International Conference on Computer Authentication System „PassPositions‟, Journal of Image and Graphics,
Applications and Information Processing Technology (CAIPT), 8-10 Vol. 6, No. 2, December 2018.
Aug. 2017, IEEE. [19] A. H. Lashkari, A. A. Manaf, M. Masrom, “A Secure Recognition Based
[15] K. Irfan, A. Anas, S. Malik, S. Amir “Text based graphical password Graphical Password by Watermarking” 11th International Conference
system to obscure shoulder surfing”, 15th International Conference on on Computer and Information Technology, IEEE, 2011.
Applied Sciences and Technology (IBCAST), 2018, IEEE. [20] Y. Higashiyama, N. Yanai, S. Okamura, T. Fujiwara, “Revisiting
[16] D. Eastlake, P. Jones, “US Secure Hash Algorithm 1 (SHA1)”, RFC Authentication with Shoulder-Surfing Resistance for Smartphones”,
Editor, ACM, 2001. Third International Symposium on Computing and Networking
[17] M. Raza, M. Iqbal, M. Sharif, W. Haider, “A Survey of Password (CANDAR), IEEE, 2015.
Attacks and Comparative Analysis on Methods for Secure
Authentication”, World Applied Sciences Journal 19 (4): 439-444, 2012
437 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Image-based object recognition is a well-studied excessive utilization of advanced technological tools, the one
topic in the field of computer vision. Features extraction for needs to access query sketch more accurately and retrieve its
hand-drawn sketch recognition and retrieval become relevant contents to be well-recognized through technological-
increasingly popular among the computer vision researchers. based smart devices. However, to acknowledge the needs of
Increasing use of touchscreens and portable devices raised the the society and to balance with technological advancement, the
challenge for computer vision community to access the sketches researchers have been analyzed various novel tasks regarding
more efficiently and effectively. In this article, a novel deep sketch recognition [5, 6], and sketch-based image retrieval [7,
convolutional neural network-based (DCNN) framework for 8], in a field of computer vision. The idea behind the sketch
hand-drawn sketch recognition, which is composed of three well-
classification or recognition is to extract the information from
known pre-trained DCNN architectures in the context of transfer
learning with global average pooling (GAP) strategy is proposed.
the desired object class of labeled sketch-images among the
First, an augmented-variants of natural images was generated pre-defined set of object-classes. Based on the extracted
and sum-up with TU-Berlin sketch images to all its information, the label of the targeted instance can be correctly
corresponding 250 sketch object categories. Second, the features identified. Classification or recognition techniques usually rely
maps were extracted by three asymmetry DCNN architectures on extracted features through instance training before making a
namely, Visual Geometric Group Network (VGGNet), Residual recognition. For sketch recognition, the researchers borrowed
Networks (ResNet) and Inception-v3 from input images. Finally, handcrafted features approaches which have been successfully
the distinct features maps were concatenated and the features used to extract features from natural images. There exist,
reductions were carried out under GAP layer. The resulting Scale-Invariant Features Transforms (SIFT) [9] Histogram of
feature vector was fed into the softmax classifier for sketch Oriented Gradients (HOG) [10], descriptors and the bag-of-
classification results. The performance of proposed framework is features has been already utilized. In this regards, different
comprehensively evaluated on augmented-variants TU-Berlin handcrafted features techniques are followed to yield the global
sketch dataset for sketch classification and retrieval task. features for sketch recognition, i.e., GF-HOG [11], FV [12],
Experimental outcomes reveal that the proposed framework SSIM [13] and Structure Tensor [14]. Usually, handcrafted
brings substantial improvements over the state-of-the-art feature representations are not considered robust and also due
methods for sketch classification and retrieval. to their high dimensionality make them computationally
expensive. Current approaches to object recognition make the
Keywords—Deep convolutional neural network; sketch
necessary use of deep learning and machine learning
recognition; transfer learning; global average pooling
techniques. However, the most existing work in sketch
I. INTRODUCTION recognition is based on deep learning approaches using deep
convolutional neural networks (DCNNs) and showed an
In a human point of view, sketch analysis is not only impressive result than handcrafted approaches [6, 8, 15].
considering a fundamental problem, but it has a prominent role
in the field of human-computer interaction (HCI). Sketches can In the recent past, deep learning frameworks based on
be seen everywhere and have a significant role in daily life DCNNs shows a breakthrough in different areas of computer
activities, i.e., education sector, art, design, and entertainment, vision, including vision recognition on large-scale challenging
etc. All through human society progress, the sketch has been dataset [16, 17]. Moreover, deep learning approaches also
utilized as a fundamental tool for conveying feelings, thoughts, benefitting sketch-based recognition and can provide useful
judgments, and opinions. Since the ancient time, the people features representations by analyzing large-scale sketch
express their views in the form of sketch related petroglyphs or dataset, such as TU-Berlin sketch benchmark [18, 19]. Deep
cave paintings. Such kind of art examples can be easily seen learning is capable of generating more distinctive features from
today in pre-historic art caves throughout the world. sketch images and can leverage the performance for sketch
classification or recognition as compared to use hand-crafted
The Technological explosion makes the sketches easy and features. Deep features for sketch recognition was first time
ubiquitous. There exist for several fascinating multimedia utilized by [20] and design a specialized neural network model.
applications such as HCI [1] and some other relevant work [2- As a result, the classification accuracy on sketch image dataset
4]. With the fame of touchscreens and smart-phone devices TU-Berlin [21] has been improved as compare to hand-crafted
encouraged the people to draw sketches digitally. Presently, an features. Similarly, two different well-known CNNs models,
*Corresponding Author.
438 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
namely LeNet [22] and AlexNet [16] are used to extract Berlin sketch dataset and also assesses on TU-Berlin
features from sketch images and show improvement in the sketch dataset (without augmented-variants).
recognition results [15]. On the other hand, some recent
attempts utilized different layers of various CNNs architectures An evaluation of the proposed framework for sketch
for features extraction for the purpose of sketch classification retrieval task.
and retrieval [8, 18, 23]. The rest of the manuscript is organized as follows: In
Visual recognition or classifications by deep learning Section 2, we briefly present related literature based on
approaches are mostly rely on extracted features. Generally, handcrafted features and deep features. Section 3 describes the
Deep CNNs features are categorized into three basic levels, overall details of the proposed approach including data
such as high, middle, and low-level features. Each level of preparation and augmentation variants used in this study, the
extracted features are having their strengths and potential in concept of transfer learning, and different proposed pre-trained
producing results and accuracy [24-26]. In order to obtain a deep CNNs architectures utilized in the current research.
higher recognition or classification accuracy and reducing the Section 4 provides results, analysis and evaluations of the
computational efforts, the concept of transfer learning (TL) proposed methodology. We conclude the manuscript in
approach can be exploited to get more robust features by Section 5 along with the future directions.
combining the learned knowledge from multiple DCNNs II. RELATED WORK
models [27, 28]. In TL approach, first, the Deep CNNs models
are trained on the generic visual dataset, and then pre-trained We include a review work for sketch recognition utilizing
models can be directly used to train on domain-specific handcrafted-feature methods. Further, we enclosed our review
datasets. The motivation behind the TL approach is to combine details about deep learning approaches which have been used
more comprehensive and relevant knowledge of input objects for hand-drawn sketch recognition and retrieval task. To hold
resulting from multiple CNN architectures and then pass them focus, we threw light on the review work entirely related to
through a classifier for a final decision. We believe that by hand-drawn sketch recognition.
doing so, it can achieve more robust and higher recognition A. Handcrafted Features
accuracy as compared to the one extracted through single deep
CNN model. Previous sketch recognition problem was handled about
CAD and artistic drawings by [32-34]. After releasing a large-
The sketches are mostly handled through smart-phones and crowed source TU-Berlin hand-drawn sketch dataset in 2012
other portable devices for different purposes in daily life by [21]. This dataset gains popularity among the computer
routines. In this regard, we attempt to facilitate such touch- vision researchers to utilize it further for recognition tasks.
screen environment to retrieve the query sketch contents with Variety of traditional approaches was carried out to classify
higher recognition rate. To overcome the existing deficiencies different categories of sketch dataset and was tried to achieve
in the sketch recognition system and following the emerging higher recognition accuracy. Some researchers employed hand-
trend of exploring deep learning for features extraction via engineered features techniques to extract the features for sketch
transfer learning approach, we proposed three different well- recognition such as scale-invariant features transforms (SIFT)
known robust DCNNs architectures in the state-of-the-art [9], histogram of oriented gradients (HOG) [10] and the bag-
visual recognition to the task for sketch recognition. The of-features techniques [35, 36].
proposed DCNNs architectures includes Inception-v3 [29] ,
ResNet [30] and VGGNet [17]. All these architectures have Although, a method proposed by [21], describe the inter-
achieved promising performance on various challenges. These class similarities and intra-class variations in large crowed-
networks are trained on large-scale image dataset ImageNet source sketch dataset. Support Vector Machine (SVM)
[31]. classifier was used to learn the sketch representation in various
object categories. Original sketch benchmark proposed by [21]
The main contributions in this manuscript can be presented and was then modified by [12]. The modified work uses SIFT,
in the following: Gaussian Mixture Model (GMM) based fisher vector encoding
for sketch recognition and fed into SVM classifier. This
A novel and efficient CNN-based framework for hand-
approach enhances the recognition performance near to human
drawn sketch recognition is proposed that exploits the
(73.1%) [21] accuracy rate against the same sketch dataset. A
strength of extracted features from the various pre-
star graph based ensemble matching strategy was employed by
trained DCNNs via transfer learning with the utilization
[37], it covers not only local feature, but global structures of
of global average pool (GAP) concept.
sketches were also adopted to match them. Further, structure
An attempt to generate the augmented-variants of matching was encapsulated, and bag-of-features was learned to
natural images paired with TU-Berlin sketch dataset for exploit in a single framework. Eitz et al. [25] demonstrated
enhancing a sketch recognition performance. hand-drawn sketch classification through implementing local
features vectors techniques, i.e., SIFT and other different
A performance analysis of three individual deep CNNs descriptors such as spark feature, shape context, HOG, and
architectures compare with proposed framework in a SHOG are embedded in a bag of features model and evaluate
context of transfer learning with GAP for sketch the performance on large scale sketch-based image dataset
recognition. The proposed framework obtained state-of- through Sketch-Based Image Recognition (SBIR) system. In
the-art recognition accuracy on augmented-variants TU- [38], the author threw light on the proposed method
Symmetric-aware Flip Invariant Sketch Histogram (SYM-
439 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
FISH) for sketch image retrieval and classification. Another through proposed model. The author [18] applied a feature
approach of multi-kernel features was demonstrated by [39], fusion approach for sketch-based recognition system to
where different local features were extracted to analyze the considered different layers of CNNs for features extraction
sketch image, integrate them to improve the sketch recognition from the TU-Berlin sketch dataset.
performance. Individually, every feature performance was
calculated and found that HOG outperformed as compare to III. PROPOSED METHOD
others. This section describes the proposed framework based on
Different researchers have made the efforts through DCNNs architectures for sketch recognition. In the proposed
handcrafted features for sketch recognition, among these, a method, three well-known Deep CNN architectures, i.e.,
Fisher vector spatial pooling (SV-SP) [12], sketch image Inception-v3 [29], ResNet [30], and VGGNet [17] in the state
representation approach raises the sketch recognition of the art of visual recognition for sketch analysis are used. The
performance up-to 68.9% to come close to 73.1% [21] human weights of these architectures are available for modification.
accuracy on TU-Berlin sketch benchmark. Generally, These pre-trained models downloaded from the webpage of
handcrafted features are not considered robust, and one of the keras [44]. The weights are loaded to all the corresponding
limitation is high dimensionality of these features make them architectures. The proposed architectures are trained on
computationally expensive. augmented-variants TU-Berlin sketch dataset. The block
diagram of the proposed framework is presented in Fig. 1.
B. Deep Features
A. Data Preparation and Augmentation-Variants
Recently, deep neural networks (DNNs) are utilized for
various kind of problems, which have shown immense To carry out this experiment, a hand-drawn sketch dataset
performance in different applications, including image TU-Berlin [21] is utilized. The learning performance of deep
recognition [16, 40, 41]. Deep networks have changed the convolutional neural networks depends on the availability of a
trend by replacing hand-engineered features to the learning large amount of training data. Data transformation and
strategy. Instead of this, a wide range of research has been deformation techniques are used for expending the training
conducted comprising natural image recognition. AlexNet [16] dataset as an additional data samples to the existing labeled
outperforms on image recognition in comparative with others, one, to reduce the overfitting problem. An essential concept of
and handle the ImageNet challenge with more significant the data augmentation is that; the labels of the instances remain
improvements. Moreover, the utilization of deep neural unchanged after applying this operation. Data augmentation
networks has been expended to other tasks with variant sizes of can improve the generalization and discriminative ability of the
network structures and depth according to the nature of the model [16].
problem. The most advanced augmentation method is adopted by
The networks, VGGNet [17], and GoogleNet [42] with mixing natural images with different transformations of
deeper structure and ability to handle the complexity enhanced edge, edge maps corresponding to the sketch images
limitations of neural networks were introduced. The emergence in the training dataset through anisotropic diffusion approach
of these deeper networks laid the foundation of a vast neural [45]. Fig. 2 illustrates natural image, edge enhanced and edge
network named ResNet [30] having the residual connection, to maps of natural image transformations. This will enable the
permit the network for identity mapping tasks between the proposed framework to compare effectively natural images; its
layers of the network. These deep neural networks were chosen various transformations i.e., edge maps and sketch images to
and exercised on natural images to overcome the problem. match for the sketch recognition task. Different
However, several deep learning approaches have been adopted transformations and mixing natural images have been used by
for sketch recognition. For the first time, an effort has been [8, 43] and extracted the features from both type of images i.e.,
made to specially design a deep convolutional neural network sketch, natural images for recognition and retrieval task. In our
(DCNN) architecture named sketch-DNN by [20]. Another case, the addition of augmentation-variants to the
research [15] extracts sketch features from two famous pre- corresponding sketch objects categories will enable the
trained CNNs, namely, AlexNet [16] and modified version of network to learn more discriminative features representations.
LeNet [22] and yield little improvement in the recognition It will also facilitate the end-user to query sketch image
results. The major contribution has been presented in [5], a through sketch retrieval system.
deep CNNs model namely Skatch-a-Net was introduced for It is stated that augmented variants with sketch images will
sketch recognition and beats the human sketch recognition enhance the generalization ability of trained CNN-based
accuracy. Later on, the existing model was modified in [6] and framework on unseen sketch images.
the sketch recognition performance gap increased from 1.8% to
4.9% than human recognition accuracy. Five convolutional Most likely, the edge-map exists in the hand-drawn sketch
layers CNN was trained by [43] by taking sketch images objects. To make it easy for the CNN-based framework to
mixed with natural images as augmented training dataset. handle the edge maps of the natural images, the enhanced
Further, to enhance the discriminative ability of the network, edges of natural images are formed and model them with edge
the training was presented with multiple rotated version of the maps of natural images. Gaussian smoothing method is utilized
sketch edge map and predicted the results with the labels. Jamil on the edge maps of natural images to form the enhanced-edge
et al. [8] attempts to recognize partially colored hand-drawn images of natural images. Mathematically, it can be
sketch images and implemented fine-tuned CNN on augmented presented as:
TU-Berlin sketch dataset to retrieve query-based sketch images
440 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Simgn Simg1 ,Simg2 ,....Simgn 1 n
(2)
where, n denoted the number of training sketch images,
similarly, for natural images;
(5)
To this end, the symbolic notations are assigned to
represent the training data of sketch images, natural images, where, n and t are the sketch and natural images with
and other augmented-variants, i.e., edge maps and enhanced augmented-variants respectively and c represents the
edges for the proposed framework. The sketch images are corresponding object category, for training the proposed
represented as: framework.
441 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Pre-Trained CNNs Architectures world, such a big amount of data is not cheap and also very
Several state-of-the-art deep neural networks (DNNs) have hard to collect. Transfer learning is the reasonable solution to
been utilized for various kind of problems and gives tackle such kind of problems and can provide an accurate result
outstanding performance in the field of computer vision with less training samples.
application such as classification, recognition, etc [46, 47]. In Transfer Learning technique is widely used in the machine
the proposed methodology for sketch recognition, three learning to utilize useful information from the set of source
different pre-trained deep CNNs models are adopted for point to the set of target point [49]. The inspiration behind the
features extraction via transfer learning, includes Inception-v3 adaptation of transfer learning is to solve a problem with
[29], ResNet [30] and VGGNet [17]. Although the improved results for the target domain. To be more specific,
architectures of these networks are different from one another, for example, the base model is first trained on relevant data
each of the adapted model architecture is describe in the instances for a specific task and then move to the target task
following: trained by their data instances [48]. Transfer learning is the
1) Inception-v3: Inception-v3[29] is a deep convolutional best choice for the case when the dataset of the source domain
is bigger than the dataset of the target domain. If the size of the
neural network and the winner of the ImageNet Large Scale
dataset for the target domain is smaller and similar to the
Visual Recognition Challenge (ILSVRC)-2014. This dataset of source domain, then overfitting possibility is high.
architecture has been implemented as an updated version of Alternatively, the chance of overfitting is reduced, and only
the GoogleNet [42]. with a depth of 44-layers and 21 million fine-tuning of the pre-trained model is required if the size of
learnable parameters. Inception module is illustrated in Fig. 3. data for the target domain is large and similar to the dataset of
2) ResNet: ResNet CNN is a very deep residual network, the source domain.
proposed by He et al.[30] to addresses the training problems
confronted by deep CNNs. This model received promising
results on ImageNet. The complexity of this network is higher
than other CNNs architectures due to the existence of its 152
layers. Shortcut connections are one of the critical innovation
of ResNet CNN, which carries out the identity mapping, and
their output is linked to stacked layers’ output. ResNet CNN is
illustrated in Fig. 4.
3) VGGNet: VGGNet is a CNN model, invented by visual
geometry group (VGG) of Oxford University [17]. This model
is the first runner-up of ILSVR-2014 for classification and the
winner of localization task. The architecture of VGGNet is
similar to AlexNet, and the only difference is the depth of the
VGGNet. This architecture consists of 19 layers, including Fig. 3. Basic Inception Module.
convolution, pooling, and three fully-connected layers. The
network consists of small convolutional kernel 3x3 with stride
1. It performs better than AlexNet. The architecture of the
VGGNet is shown in Fig. 5.
C. Transfer Learning
In the context of traditional machine learning algorithms, it
is assumed that the characteristics of features space based on
training and testing data are equal [48]. However, in a practical Fig. 4. Basic Architecture of ResNet.
442 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
443 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
100
Fig. 8. Sample Colored Natural Images with Corresponding Sketch Image.
Accuracy (%)
60
classification accuracies are shown in Table I. According to
tabulated outcomes, the performance of proposed individual
40
CNN architectures, i.e., VGGNet, ResNet, and Inception-v3
obtained 78.93%, 89.61%, and 91.89% classification accuracy,
respectively. However, the proposed framework achieves 20
better performance with 94.57% sketch classification accuracy
on augmented-variants TU-Berlin sketch dataset and beat the
0
individuals, i.e., VGGNet, ResNet and Inception-v3 in
VM LE M -S
P N D
performance gap with 15.64%, 4.96%, and 2.68%, -S MB -SV FV MA T HO
G SE MK HU ME
HO EN R
respectively. It is evident from the tabulated results that OU
proposed method outperforms in terms of sketch classification Fig. 9. Sketch Classification Accuracy based-on Hand-Crafted Features.
than the performance of other three individual architectures.
TABLE. I. COMPARATIVE SKETCH CLASSIFICATION ACCURACY OF THE
In this subsection, the performance of proposed framework PROPOSED FRAMEWORK WITH OTHER INDIVIDUAL DCNNS ARCHITECTURES
is compared with state-of-the-art methods including sketch-
based handcrafted features and deep features methods. The Index Method Accuracy
overall accuracy is shown in terms of percentage to 1 VGGNet 78.93%
demonstrate the results. Fig. 9 shows the hand-crafted features
recognition performances on sketch images. The proposed 2 ResNet 89.61%
method outperforms on HOG-SVM (recognition accuracy 3 Inception-v3 91.89%
56.0%), sketch recognition accuracy achieved through 4 Proposed Method with GAP 94.57%
Ensemble method, Multi-kernel-SVM and Fisher Vector-SP
were 61.5%, 65.8%, and 68.9%, respectively in the literature Therefore, some conclusions can be made from the baseline
study. results mentioned in the Fig. 9 and Table II. First of all, the
HOG based features with SVM classifier has the lowest proposed CNNs-based framework consistently outperforms in
recognition rate. However, the best performance accuracy sketch classification on both handcrafted features methods and
based on hand-crafted features is 68.9%, which is less than the the sketch features analyzed through deep neural network
human recognition accuracy (73.1%) on sketch data. The models. This can show that the use of natural images plays a
reason of lower performance of handcrafted features is that; significant role in the evaluation of sketch images.
mostly these methods have been designed to extract features Additionally, various augmentation-variants, specifically edge
from real photos and not suitable to covers the high variability maps strengthen the proposed framework capability in sketch
of abstractions and appearance in sketch images. recognition accuracy. Secondly, the performance of individual
deep CNNs becomes improved when it goes deeper. The best
Similarly, the results for deep features methods are recognition performance through individual deep CNN
summarized in Table II. It shows that deep networks perform architecture is achieved by Inception-v3. Thirdly, in a case of
better than hand-crafted features. Human recognition level transfer learning approach, the distinct features of three deep
accuracy on TU-Berlin dataset was first beats through deep CNNs architectures were combined, and the recognition
network architecture [5] and enhance recognition rate with performance is improved by employing GAP strategy. This
1.8% higher than human recognition performance. Moreover, outcome substantiates the experiments for augmented-variants
the performance gap grows from 1.8% to 4.2% when [43] TU-Berlin sketch dataset. In the proposed framework, it is
implemented deep sketch model by mixing sketch images with declared that using transfer learning with GAP increases sketch
colored images for sketch recognition. The sketch recognition recognition accuracy. This is also applicable for combining
accuracy 79.1% achieved by [8] using pre-trained VGGNet distinctly extracted features from multiple CNNs architectures
architecture through transfer learning approach. as compared to individual CNN architecture.
444 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. SKETCH CLASSIFICATION ACCURACY COMPARISON BASED ON On the other hand, we evaluate our proposed method on
DEEP FEATURES
TU-Berlin sketch dataset (without augmented-variants), the
Index Methods Accuracy experimental results are illustrated in Fig. 10. The proposed
framework achieves a competitive performance of 72.82%
1 AlexNet-SVM [16] 67.1%
classification accuracy as compared with 73.1% human
2 AlexNet-Sketch [16] 68.6% recognition accuracy except those sketch based CNNs
3 LeNet [22] 55.2% architectures [5, 7], which have been specifically designed for
sketch classification.
4 Human [21] 73.1%
5 Sketch-a-Net [5] 74.9% E. Further Evaluation for Retrieval Task
6 Deepsketch [43] 77.3%
The performance of the proposed deep framework is further
evaluated on sketch retrieval task. For this test, the proposed
7 VGG-based Transfer Learning [8] 79.1% deep CNNs-based frame work is used to extract features from
8 Proposed Method with GAP 94.57% both the sketch images and natural images. All the images are
indexed with concerned features. For retrieval task, proposed
framework is used to extract the features from the edge maps
and query sketches separately and compared with all the
retrieval candidate images in the database. Euclidean distance
is computed to make the comparison between the query sketch
images and the images in the database. The query images are
randomly selected to retrieve the similar images from the
image database.
The sketch object-based queries and top-9 retrieval results
are shown in Fig. 11. The retrieval results are ranked with
scored values. The lower scored value represents the higher
rank similarity between the query sketch and the retrieved
image. In most of the cases, the query images retrieved the
most similar candidate images which show the enough
discriminative features for retrieval task.
These images (i.e., teapot, beer-mug, guitar, etc.) are
retrieved with high rank similarity. Interestingly, the retrieved
image had very comparable edge maps which make the
retrieval task in high ranks. Moreover, in some cases the
system failed to retrieve the right candidate images, the reason
might be the natural images having complex background
Fig. 10. Sketch Classification Results Comparison on TU-Berlin (without leading the large difference between the sketch images and
Augmented Variants). edge-maps.
Fig. 11. Retrieval Performance of our SBIR, Top-9 Retrieval Outcomes of Four Sketch Queries by Proposed CNNs-based Framework. The First Column Indicates
the Query Sketches, and the Most Similar Retrieved Candidate Images According to Ranked-Score from Four Different Object Categories are Shown From the Top
to the Bottom-Row in the Sequence. Red-Box Images Indicate Incorrect Retrieved Images.
445 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The capabilities of the proposed framework are extensively augmented-variants are beneficial for sketch recognition. It is
evaluated on a retrieval task. The lower rank similarity retrieval worth mentioning that handcrafted features performance on
results with the corresponding scored values are illustrated in this dataset is worse than deep feature methods. In our case, the
Fig. 12. Additionally, the retrieval performance of the proposed complex background of the natural images and generated edge
framework was also tested on sketches and natural images maps from those images, make the proposed method more
which was not a part of training or validation data. It is challenging and competitive. The proposed method performed
interesting to demonstrate that the query sketches retrieve the better and retrieved the candidate images mostly in high ranked
nearest candidate images with less ambiguity. However, the score of similarity. This demonstrate that proposed framework
performance can be further improved by providing enough based on transfer learning with GAP is capable to extract the
training data instances and also by reducing background most discriminative features from both type of images i.e.,
complexity of the natural images. Results illustrated in Fig. 13 sketch and natural images and could help to strengthen
advocate that proposed CNN-based framework is capable to retrieval performance. But Fig. 12 and Fig. 13 represent the
perform well on the variety of images which was not a part of retrieval results of lower rank similarity and the results for
training or validation data. These experiments validate the images instead of using training and validation images
effectiveness of the proposed framework for sketch respectively, where the incorrect retrieved images are outlined
recognition. in red-boxes. Therefore, it is stated that the lower rank
similarity performance and incorrect retrieved images might be
F. Experimental Analysis the reasons of providing not enough training samples to the
The proposed CNNs-based framework outperforms all proposed framework as well as it might be not well-aligned
baselines and achieves better performance on TU-Berlin with complex background of candidate retrieval images.
(augmented-variants) sketch dataset which shows that
Fig. 12. Low-Rank Similarity Retrieval Performance of SBIR. The Red-Outlined Boxes Indicate Incorrect Retrieved Candidate Objects.
Fig. 13. Retrieval Performance based on Sketch-Query for other Object Categories.
446 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
447 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[36] T. Joachims, "Text categorization with support vector machines: [45] P. Perona, and J. Malik, “Scale-space and edge detection using
Learning with many relevant features." pp. 137-142. anisotropic diffusion,” IEEE Transactions on Pattern Analysis and
[37] Y. Li, Y.-Z. Song, and S. Gong, "Sketch Recognition by Ensemble Machine Intelligence, vol. 12, no. 7, pp. 629-639, 1990.
Matching of Structured Features." p. 2. [46] M. E. Paoletti, J. M. Haut, R. Fernandez-Beltran, J. Plaza, A. J. Plaza,
[38] X. Cao, H. Zhang, S. Liu, X. Guo, and L. Lin, "Sym-fish: A symmetry- and F. Pla, “Deep Pyramidal Residual Networks for Spectral–Spatial
aware flip invariant sketch histogram shape descriptor." pp. 313-320. Hyperspectral Image Classification,” IEEE Transactions on Geoscience
and Remote Sensing, vol. 57, no. 2, pp. 740-754, 2019.
[39] Y. Li, T. M. Hospedales, Y.-Z. Song, and S. Gong, “Free-hand sketch
recognition by multi-kernel feature learning,” Computer Vision and [47] J. Kim, B. Kim, P. P. Roy, and D. Jeong, “Efficient Facial Expression
Image Understanding, vol. 137, pp. 1-11, 2015. Recognition Algorithm Based on Hierarchical Deep Neural Network
Structure,” IEEE Access, vol. 7, pp. 41273-41285, 2019.
[40] N. Mboga, S. Georganos, T. Grippa, M. Lennert, S. Vanhuysse, and E.
Wolff, “Fully Convolutional Networks and Geographic Object-Based [48] K. Weiss, T. M. Khoshgoftaar, and D. Wang, “A survey of transfer
Image Analysis for the Classification of VHR Imagery,” Remote learning,” Journal of Big Data, vol. 3, no. 1, pp. 9, May 28, 2016.
Sensing, vol. 11, no. 5, pp. 597, 2019. [49] S. J. Pan, and Q. Yang, “A Survey on Transfer Learning,” IEEE
[41] D. Mao, and Z. Hao, “A Novel Sketch-Based Three-Dimensional Shape Transactions on Knowledge and Data Engineering, vol. 22, no. 10, pp.
Retrieval Method Using Multi-View Convolutional Neural Network,” 1345-1359, 2010.
Symmetry, vol. 11, no. 5, pp. 703, 2019. [50] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z.
[42] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-
Erhan, V. Vanhoucke, and A. Rabinovich, "Going Deeper with Fei, “ImageNet Large Scale Visual Recognition Challenge,”
Convolutions," arXiv e-prints, https://ui.adsabs.harvard.edu/abs/ International Journal of Computer Vision, vol. 115, no. 3, pp. 211-252,
2014arXiv1409.4842S, [September 01, 2014, 2014]. December 01, 2015.
[43] X. Wang, X. Duan, and X. Bai, “Deep sketch feature for cross-domain [51] M. Lin, Q. Chen, and S. Yan, "Network In Network," arXiv e-prints,
image retrieval,” Neurocomput., vol. 207, no. C, pp. 387-397, 2016. https://ui.adsabs.harvard.edu/abs/2013arXiv1312.4400L, [December 01,
2013, 2013].
[44] “https://keras.io/applications/.”
[52] “https://www.kaggle.com/jessicali9530/caltech256 ”.
448 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—All over the world, many research studies focus on transport networks, road traffic density and daily requests for
developing and enhancing real-time communications between parking space.
various transport stakeholders in urban environments. Such
motivation can be justified by the growing importance of Several studies examine the possibility of establishing both
pollution caused by the transport sector in urban areas. In this multimodal transport networks models, road states and the
work, we propose an approach of assistance for displacement in availability of car parks in order to propose approaches that
urban environment taking advantages of multimodal urban calculate and determine the most appropriate paths for a user’s
transportation means, where several modes of public transports request. These approaches must be robust enough to keep their
are available. In addition, we also consider the possibility of using efficiency and effectiveness as the complexity of the
both private modes of transport and cities parking. The proposed multimodal network and the amount of data to be processed
distributed approach described in this paper is based on an increases.
abstraction of a city multimodal graph according to the available
modes of public transport and road traffic and transition graph The major issues in the development of these approaches
approach to move from a mode to the other mode. Numerical can then be considered in two aspects:
results are developed to justify the effectiveness of our approach.
1) Management of heterogeneous data: Different modes
Keywords—Multimodal transport; distributed approach; correspond to a set of navigation datasets which are acquired,
transition graph stored and managed by different public or private
organizations. Likewise, car parks management systems make
I. INTRODUCTION it possible to retrieve data concerning the availability of car
Displacements in urban environments are one of the major parks at a given moment and to predict the occupancy rate for a
problems facing most supercities around the world. In addition later date. By the same token, road traffic supervision systems
to the pollution and time waste generated by traffic jams provide real time data not only from traffic roads states but also
expectations while looking-up for parking spots to park private a full history of occupancy rate related to road traffic arcs [5].
vehicles [1], urban mobility favors certain health disturbances 2) Multimodal network and optimal path finding
among cities citizens. In recent year authorities have
approach: Different types of data should be considered in
encouraged the use of a diversity of transport modes in order to
reduce the impact of this problem [2]. determining the optimal path in a multimodal network. Various
works then seek to elaborate the problem modeling approach
Several solutions have then been developed to help users to which allows an adequate structuration and an easy
choose the optimal path to follow to reach their destinations. accessibility to the information which is necessary for different
Early solutions concerned each a particular mode of public
computations [6]. Likewise, the resolution approach must also
transport and each operator proposed a system that manages its
own network. Subsequently, other solutions made it possible to be sufficiently powerful and flexible to be able to maintain its
propose the best optimal paths by combining two or more performance when increasing the size of the multimodal
modes of public transports [3] [4]. network [7].
Currently, researches are moving towards the proposal of In the present paper we are going to present a new model of
an urban mobility management system by considering both all the multimodal network based on the graph theory and a
available cities public modes of transports, traffic road states parallel distributed approach to propose optimal paths between
and the availability of city car parks. Thus, the optimal route two points in a city.
for a user would be to drive his private vehicle to a parking lot, The rest of the paper will be developed as follows: in
then take a series of public transport lines. Section II we present definitions of the concepts necessary for
In order to be able to ally with the development of the city's the description of the problem and the proposed solution.
urban transport infrastructure, urban travel assistance services These definitions concern the modeling of the multimodal
must be able to consider the growing complexity of public transport network using graph theory. Then we proceed in the
third part to the mathematical modeling of the problems in
449 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
order to propose the approaches of resolution allowing at first An edge defined by expresses that it is
to reduce the complexity of the problems and then to propose possible to go from node to node using transport
the optimal solutions. We conclude our work with a conclusion mode . A value is associated to each edge
and perspectives.
indicating the cost of the edge at departure time according
II. PROBLEM FORMULATION to the criterion (e.g., distance or duration).
In this section, we present some necessary definitions for A. Définition1: Multimodal Path
the formulation of the optimal path problem in a multimodal
Given a multimodal transport graph . A
transport network using graph theory approach. The
particularity of the problem is that the multimodal transport multimodal path is a sequence of possible edges to go
network contains several subnetworks, each subnetwork refers from the node to the node ,
to either a particular mode of transport or the road traffic ( ).
network of the city.
The model should propose an optimal structuring of the B. Définition2: Cost Function
data in such a way that the system could have rapid access to The vector-valued function:
each arc at any subnetwork of the multimodal network. Then it represents the cost of the path departing at time according
would be able to react in real time on following any change to the criteria . is a set of vectors, where each vector
that may happen on an arc of the multimodal network. represents a criterion.
Different models have been proposed to consider the dynamic
C. Definition 3: Optimal Path Problem
particularity mentioned above, generally based on graph
theory: hypergraph theory [8] [9] [10], hierarchical graph [11], Given a graph , the optimal path problem
colored graph approach [12] and transfer graph approach [13]. according to the criteria consists in calculating a path
However, most models fail to consider the following three from node to node departing at where is
components of a city's multimodal network simultaneously: minimal. This is called the optimal path (OP).
public transport modes, road traffic network and city car parks
[14]. D. Transition Graph
Given a multimodal graph , the transition
In our study, we consider a global multimodal network graph is defined as where is
composed of: a set of monomodal public transport networks, a
set of parking spaces and a traffic road network. Each the set of monomodal graphs, and is the set of virtual
component of the multimodal network has its own specific transition edges which interconnect them. Each component
parameters. is such that ⋃ ,
⋃ , ⋃ .
So, in regard to the resolution approach, the procedure of
calculating the shortest path should be able to keep its The transition graph model consists in abstracting different
efficiency and computing power when changing the size of the modes of transport on the same map; we can distinguish in a
multimodal network; In most studies, the optimal path transfer graph between two groups of paths: intra-
resolution approach in a multimodal network is considered as a components paths and inter-components paths. An inter-
whole. Thus, to perform calculations, the approach must component path refers to a path that connects two nodes
traverse all nodes and arcs of the said network, which slows belonging to distinct modes of transport [17], while an intra-
down the computation process and reduces the system component path defines a path that links two nodes belonging
performances. to the same mode of transport
Wherefore, in the approach, the calculations are done in a Figure 1 illustrates an example of a transfer graph, where
parallel and distributed way for each component of the global Mode 1, Mode 2 and Mode 3 represent three modes of public
network. Thus, to calculate the optimal path between two transport and Road Graph represent the road network.
points of a multimodal network, our approach consists on distinguishes between two categories of monomodal graphs
making intermediate calculations for each sub graph before can be described as follow:
considering the whole multimodal network. This need to
reduce the size of the graph is justified by the complexity of the Mode Graph represent public modes (Mode 1, Mode 2,
Mode 3): nodes represent stations of the mode and
problem. Thus, with a graph of nodes, edges and modes
edges represent the paths of the transport mode
of transport, the corresponding graph generated will have a
(Tramway, bus, subway ...). Parameters of the graphs
complexity in at least of [15] [16].
(departure times at a node, estimated duration to travel
In the following, we define some concepts that will allow to an arc ...) depend on the mode of transport and are
present our model and our approach of resolution: given by the public transport operator. These
parameters can be updated by a user or an operator
Let denotes a multimodal transport graph, once a disturbance occurs on a line of the mode.
where { } is a set of nodes belonging to the
mode , is a set of modes, Road Graph represents the network of the road traffic:
is a set of packing spaces. local nodes represent the intersections of the paths and
transfer nodes represent car parks.
450 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A citizen can drive his own car from an initial point to a : logical variable defines whether the edge
second point, then park his car in a parking lot and finish his linking a node belonging to a mode to a node j belonging
travel with a public mode of transport. Different parameters to a mode n is used or not.
related to the state of an arc of a road network are updated by
users who browse it in real time. ( ): cost function of the direct arc linking a node
belonging to a mode to a node j belonging to a mode n
In the current study, transfer node represents walking
departing at time .
movements either to go from a parking space to a public
transport mode (if the latest mode of transport is a private The optimal global function should consider all possible
vehicle), or to go from a station of a mode of transport to a criterions and proposes the set of optimal paths according to all
station of another mode of transport. criterions.
III. PROBLEM RESOLUTION AND RESULTS In our study, for problem simplification reasons, we
consider that the optimal global function is a direct weighting
The objective function is a multi-objective function. of optimal paths according to each criterion [18].
Indeed, by varying the multimodal path between two nodes of
the multimodal graph, various parameters can vary (duration, The minimization of the global function can be written as
cost, level of comfort ...). follow:
Let be a multimodal path from node to node ∑ (2)
. The optimal path between a node belonging to a mode Where are weighting coefficients and refers to a user’s
to a node belonging to a mode according to the criteria preference according to the criteria .
can be defined as follow:
The optimization of the global function is subject to the
∑ ( ) (1) following constraints:
Where:
451 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
452 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
,
1200.00
1000.00
Mean Execution Time (s)
800.00
600.00
400.00
200.00
0.00
50 100 500 1000 5000 10000
Number of nodes of the Multimodal Network
453 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[13] H. Ayed, C. Galvez-Fernandez, Z. Habbas and D. Khadraoui, "Solving [17] Z. Jianwei, L. Feixiong, A. Theo and T. Harry, "A multimodal transport
time-dependent multimodal transport problems using a transfer graph network model for advanced traveler information systems," in Procedia -
model," Computers & Industrial Engineering, vol. 61, no. 2, pp. 391-401, Social and Behavioral Sciences, 2011.
2011. [18] D. Kalyanmoy, "Multi-objective Optimization," in Burke E., Kendall G.
[14] D. M. Sergio and R. Silvia, "An architecture for a Mobility (eds) Search Methodologies. , Boston, Springer, 2013, pp. 403-449.
Recommander System," Procedia Computer Sscience, vol. 98, pp. 425- [19] F. Reza Zanjirani, M. Elnaz, S. W.Y and . R. Hannaneh, "A review of
430, 2016. urban transportation network design problems," European Journal of
[15] Z. Athanasios and W. Whitney , "An intermodal optimum path algorithm Operational Research, vol. 229, no. 1, pp. 281-302, 2013.
for multimodal networks with dynamic arc travel times and switching [20] S. Juliana Verga , Y. Akebo , C. S. Ricardo and S. Wesley Vagner Inês ,
delays," European Journal of Operational Research, vol. 125, no. 3, pp. "Urban Transport and Traffic Systems: An Approach to the Shortest Path
486-502, 2000. Problem and Network Flow Through Colored Graphs," in Nazário
[16] T. Gräbener, "Calcul d'itinéraire multimodal et multiobjectif en milieu Coelho V., Machado Coelho I., A.Oliveira T., Ochi L. (eds) Smart and
urbain: Modélisation et simulation," Université des Sciences Sociales, Digital Cities. Urban Computing. Springer,, Cham, 2019.
Toulouse I, France, 2010. [21] C. Xinyuan and K. Inhi, "Modelling Rail-Based Park and Ride with
Environmental Constraints in a Multimodal Transport Network," Journal
of Advanced Transportation , vol. 2018, pp. 15 pages,, 2018.
454 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Processing speed and memory recall ability are two The implications are for more reaching. A lot of small scale
major Human Capital Intelligence attributes required for business ventures and organizations had packed up due to
recruitment. Matzel identified five domains of Intelligence. incompetent labour force. Experts are being sought from
Unfortunately, there were no stated means for measuring them. developed countries for staff retraining and subsequent
This paper presents a framework for measuring Processing speed salvaging of ventures. Output from companies continues to
and Memory intelligence domains using Sternberg and Posner dwindle in quality and number due to incompetent labour
paradigms of short memory scanning test. A Semi-Latin square force. Business firms, institutions and the society at large
was constructed and used as a competitive platform for n= 20 continue to lack in competent and skilled labour force. The
student-applicant contestants. The Cumulative Grade Point
existing recruitment process is not producing the desired
Average rankings of 20 randomly selected final year student-
applicants were used for the test. Results show that the CGPA
competent labour that is needed to contend with the fast
performance ranking of the student-applicants differ from that increasing economic challenges and production requirements
of the HCI using the framework. A Wilcoxon Signed-Ranks Test because it solely depends on certificates obtained from sick
was used to determine if the disparity in performance ranking education system. The existing recruitment attempts at
was significant. Results show that there is indeed a significant ascertaining proficiency relies on human resource tools which
difference in the performance ranking of the student-applicants includes aptitude tests (either written or computer based) and
using both approaches. The automated Construct was basically oral interviews. These methods come with their short
implemented using PHP and Mysql and deployed at falls. A person’s score in an aptitude test is known to be a
(hcipredictor.eu3.org). function of the examiner, the subject matter focus and the
educational background. These short falls raise doubts on
Keywords—Memory recall ability; processing speed; Sternberg issues like possibility of test questions being revealed by the
paradigm; Posner paradigm; human capital intelligence; Semi- examiners before the test, possibility of favoritism during oral
Latin square interviews and other related issues. These challenges continue
to make it difficult to select competent labour force with sound
I. INTRODUCTION
human capital Intelligence value especially in developing
Human capital intelligence (HCI) is an embodiment of countries. One way to improve the quality of labour force is to
knowledge, creativity, talents, habits, social and personality focus on means of measuring Intelligence as a human capital
attributes, inherent in a person which could be measured in value. Intelligence has been identified as a major index in
terms of economic value. HCI is a major factor to be recruitment. Intelligence is not only related to the extent of
considered especially for labour recruitment, leadership knowledge gained or acquired by the individuals. It reveals the
positions and managerial posts. Choosing the appropriate capability to yield from proper training, reason conceptually,
labour force for a business venture, leadership ability or think and solve problems [2]. Intelligence varies among
managerial position has always been an uphill task especially individuals hence the need for Intelligence tests. Intelligence
in developing countries. This is so because the emphasis and tests are known to specifically measure abilities of a person
major criteria for choosing the labour force is based on paper while cognitive tests measure a person’s learning in a specific
qualification and certification of the applicants. This leads to subject area. Intelligence tests are known to produce desired
erroneous judgment in the quality of labour employed. This is results and high predictive values and when it is combined with
evidently true because the quality of education in such well-structured interview it could have the highest predictive
countries is questionable. The education system is value of all the methods of selection [3]. The aim of this paper
characterized by untested facilitators, examination malpractice, therefore is to proffer a standard method of measuring human
inconsistent education policies and poor funding and a lot capital intelligence of applicants for recruitment into
more. A Nigerian education critic decries the condition of her establishments and industries rather than depending on their
educational system which gets bedeviled by the day as people over-rated certificates. The sole objective is to ensure that the
are no longer judged by the latent ability in them but the measurement is done under an equitable and competitive
certificates they have gathered by whatever means [1]. As a platform. The measurement is achieved using a short memory
result of these, education system of developing countries test which has speed and memory recall abilities as its
continues to produce half-baked graduates with low HCI value. yardstick.
*Corresponding Authors.
455 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The major aim of the research therefore is to present a whether the test digit is contained in a short memorized
framework for measurement of processing speed and memory sequence of digits previously displayed. In this manner, high
intelligence domains as HCI parameter which was lacking in speed scanning ability in human memory could be determined
[4]. The specific objective is to test the developed framework per individual and could be used as a measure of intelligence.
using 20 graduating student-applicants of the department of Similarly, human intelligence and memory recall ability has
Computer science, University of Nigeria, Nsukka. The research been measured using Posner approach. The task presents
question is interested in finding out if the results from the participants with pairs of uppercase, lowercase, and mixed-case
existing approach are the same as that of the proposed system letters (drawn from the set A, a, B, b) side by side, 0.5 cm apart
under the hypothesis that: on the screen, and these participants were asked to determine,
as quickly as possible, whether the letters were the same or
H0: The results from both approaches are the same different according to a particular rule. Participants indicated
Vs that the letters were the same by pressing the M key on a
standard keyboard and different by pressing the Z key [14].
H1: The results from both approaches differ These two methods of measuring human intelligence have
The significance of the work is to use the developed proved successful and had been the basis of research for many
framework to especially aid human resource and recruiting years. It has also been applied to animals in successfully
agencies in recruiting more qualified labour that will enhance measure of psychometric intelligence and reaction times in
production. pigeons [15]. Unfortunately, these paradigms were applied
solely for either Speed or memory recall ability. In other
II. RESEARCH BASIS words, the measurement does not take into cognisance other
According to [5], Intelligence could be measured under domains of Intelligence. Secondly, the test is a onetime effort
three basic abilities: creative, analytical, and practical abilities. which may not really reflect the true ability of the participant in
Defining Information processing in terms of creative question. There is also no active interactive competition among
Intelligence, then we will see it as the ability to convert latent the participants. Semi-Latin square presents a perfect platform
information into manifest information [6]. If we also consider for competition among participants whose human capital
that in real time, information processing requires instant intelligence value is to be measured. Semi-Latin squares have
decisions within limited time, we will also see speed and found application in many areas of life. In Agricultural
perception as creative and analytical intelligence respectively. experiments for instance, the work of [16] gives credence to
According to [4], five domains of Intelligence has been the use of a special group of Semi-Latin square known as
identified as Reasoning domain, Processing speed domain, Trojan squares as an experimental design.
Memory domain, Comprehension domain and an Unknown III. THEORETICAL BACKGROUND
domain which may be in existence. These identified domains
reduce to specific ability tests such as reasoning tests, speed A. Semi-Latin Square for Equitable and Competitive Platform
tests, memory tests and spatial tests. The challenge has always According to [17], an (n × n)/k Semi-Latin square is an n ×
been how to proficiently measure these abilities distinctly. For n array containing nk letters in such a way that each row-
instance, attempts at measuring the memory recall ability has column intersection contains k letters and each letter occurs
been and is still ongoing. According to the authors in [7], [8], once in each row and once in each column. It suffices to say
[9] and [10], several attempts had been made at measuring that no letter occurs more than once in each row and in each
memory using mathematical models such as SAM, REM, column where they are found. There exists a special type of
MINERVA 2 model, EEG analysis and ERP. Although Semi-Latin square called Trojan Square. A Trojan square is
foundational research on memory argues that short term simply an arrangement obtained by superposition of k mutually
memory differ from long term memory which presumes that orthogonal n × n Latin square (where such square exists),
their method of measurement should also differ. However, involving k disjoint sets of n varieties so that the resulting
recent unified attempt at measuring memory ability is seen in square has kn varieties, each occurring in n experimental units,
[11] and [12] where EEG, FMRI studies and Serial Recall n rows and n columns, with each row intersecting each column
Paradigms. It is not therefore the difference in memory ability in a block of k experimental units. Trojan squares are
that matters but the mode of measurement. Most of the existing constructed by superposition of two mutually orthogonal Latin
measuring procedures attempt to measure these distinct squares. Trojan squares are known to be A-, D- and E-optimal
domains using a generalization approach which arguably does among all binary incomplete-block designs of the same size
not address the peculiar nature of these domains. For instance, [17]. The optimality feature of Trojan squares gives credence
Human intelligence and memory recall ability has been for using them in developing the competitive platform for
measured in literature using the Sternberg paradigm [13]. assessing contestants.
Sternberg information processing is an information processing
paradigm that tests an individual recalling ability. It thrives to B. Construction of Semi-Latin Squares
ascertain intelligence ability of humans to scan the memory in A Semi-Latin square is constructed by superposition of the
high speed for information retrieval. The Sternberg experiment Latin squares for instance, given two Latin squares 1 and 2, a
involves different trials of experiment in which a random series semi-Latin square is obtained by superimposition of Latin
of say from one to six different digits are displayed at fixed squares as shown in Fig. 1.
point on the screen for 1.2 seconds delay time. Also given a
test digit after a 2 seconds delay time, subjects are to judge
456 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Latin Square 1 Latin Square 2 Semi-Latin Square A. Construction of Semi-Latin Square for the Competitive
Platform
The twenty students were first divided into four groups of
five (5) students each. Using only their user names and
designated symbols, the layout is shown in Table I and
Table II.
Fig. 1. A (3 × 3)/2 Semi-Latin Square.
The four groups gave rise to four sets of mutually
A B orthogonal Latin squares which were constructed using the
following equation [18].
(1)
C D
The four Latin squares are then superimposed together to
get the (5×5)/4 Trojan square in Table III.
Fig. 2. (3×3)/2 Bipartite Variety Concurrence Graph. OFFOR (1) GIFT (A) NGO (a) OKPA (@)
The bipartite variety concurrence graph of the semi-Latin UGWU (2) KACH (B) EZE (b) CHI (#)
square is also shown in Fig. 2.
C. Modified Sternberg and Posner Paradigm for Analytical EMMA(3) VIN (C) OBI (c) UCHE ($)
and Creative Intelligence
JOHN(4) UWA (D) ROSE (D) KALU (%)
The Sternberg paradigm for analytical test of human
intelligence tests for the ability and speed to recall and give
answers to analytical problems. The HCI uses a modified IKE(5) AGHA (E) OKO (e) JOEL (&)
version of the paradigm to test for analytical intelligence. The
Sternberg paradigm in [12] was modified by displaying TABLE. II. LAYOUT OF THE COMPETITIVE PLATFORM
computer simulated arithmetic expressions requiring each
contestant to complete under 5 seconds e.g. 5 + [ ] = -2. Latin 1 Latin 2
Similarly, a selection of pictures ranging from household 1 2 3 4 5 A B C D E
items, fruits, human body parts and animals were randomly
displayed in an inverted mode for contestants to quickly 2 3 4 5 1 D E A B C
identify in less than 5 seconds. Similarly, the Posner paradigm 3 4 5 1 2 B C D E A
was also modified to test for creative intelligence. The Posner
4 5 1 2 3 E A B C D
paradigm was also modified by asking the contestants to
identify missing vowels in certain words and displaying some 5 1 2 3 4 C D E A B
reversed words for the contestants to identify the word within 5 Latin 3 Latin 3
seconds. The words used in this expert system are less than or
equal to 5 in length and are obtained from advanced learners a b c d e @ # $ % &
dictionary. c d e a b & @ # $ %
457 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
458 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
CGPA Ranking
HCI Ranking
CGPA score
User name
HCI Score
Deviation
CGPA
TR
ID
Fig. 4. Sternberg Arithmetic Test for Analytical Intelligence.
Using the competitive Semi-Latin square platform show in 2 OFOR 4.65 93.0 93.00 5 1 2 1
Table III, the Posner and Sternberg test for memory and speed 3 UGWU 2.52 50.4 15.75 16 16 13 3
processing was carried out by the developed application. Each 4 EMMA 4.59 91.8 91.80 5 2 1 1
scheduled group of four (4) competing applicants is meant to
answer twenty (20) questions. The performance ranking of 5 JOHN 3.30 66.0 23.57 14 12 17 5
each applicant is noted for each competition. In this case, there 6 AGHA 2.77 55.4 15.11 18 14 16 2
were five (5) competition schedules which gave five (5) 7 VIN 1.69 33.8 9.70 20 20 18 2
different performance ranking and another five (5) different
sub total score for each applicant. The scaled CGPA score 8 GIFT 2.29 45.8 12.05 19 17 10 3
represents the contestant’s score based on the Senate approved 9 UWAH 3.91 78.2 32.58 12 9 9 0
CGPA. The scaled CGPA is computed using the following 10 KACH 2.08 41.6 10.40 20 19 19 0
equation:
11 OKO 4.03 80.6 57.57 7 4 5 1
(3) 12 EZE 2.18 43.6 12.11 18 18 15 3
The HCI score is obtained by taking into cognizance the 13 ROSE 4.01 80.2 57.29 7 5 4 1
total score for each student in all the sessions and their 14 NGO 4.31 86.2 61.57 7 3 3 0
respective total rank score (TR). The result is then scaled to
15 OBI 3.45 69.0 28.75 12 11 14 2
100 percent. The HCI score is obtained using:
16 JOEL 2.74 54.8 19.57 14 15 20 5
( ) (4) 17 KALU 3.49 69.8 31.73 11 10 12 2
Where TQ = total questions, TR represents the total 18 UCHE 3.93 78.6 39.30 10 7 7 0
performance ranking of each student based on their position 19 CHI 4.01 80.2 40.10 10 6 6 0
after every successful group competition. 20 OKPA 3.68 73.6 33.45 11 8 8 0
For instance, IKE has a CGPA score = 56.6, TR = 14, TQ =
100, n = 5. The HCI score will thus be calculated as:
( ) = 20.21
459 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
460 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[6] G.W Jeroen and M.S. Richard, “Models of Memory”in Stevens”, [12] S. Sternberg. “High-Speed Scanning in Human Memory”, Science, New
Handbook of Experimental Psychology, Third Edition, 1988, Volume 2: Series, Vol.153, No. 3736. Pp. 652-654, August, 1966. Made Available
Memory and Cognitive Processes. (Pp.43-76). New York: by JSTOR on September, 2005 Retrieved from:http://www.jstor.org/.
JohnWiley&Sons,Inc. [13] A. Douglas and F. Bert “Age, Speed of Information Processing, Recall,
[7] H.U. Amin, A.S Malik., N. Badruddin and,W.T Chooi “ Brain Behavior and Fluid Intelligence”. University of Toronto, Scarborough Campus,
in Learning and Memory Recall Process: A High-Resolution EEG Ontario, Canada., pg. 229-248, 1995.echnical Writer’s Handbook. Mill
Analysis. In: Goh J. (eds) The 15th International Conference on Valley, CA: University Science, 1989.
Biomedical Engineering. IFMBE Proceedings, vol 43. Springer, 2014. [14] C. N. Aljoscha, R. Rainer, M. Ralf and A. Alois “Intelligece and
[8] H.U Amin, A.S. Malik, S. Aamir, N. Kamel, W.T Chooi and Reaction Times in the Hick, Sternberg and Posner Paradigms”. Person.
H.Muhammad. “P300 correlates with learning & memory abilities and individ. Diff. 1997, Vol. 22, No. 6, pp. 885-894ouglas and F. Bert
fluid intelligence”, Journal of NeuroEngineering and Rehabilitation, Vol “Age, Speed of Information Processing, Recall, and Fluid Intelligence”.
12, Article number: 87, 23rd September, 2015. University of Toronto, Scarborough Campus, Ontario, Canada., pg. 229-
[9] S. Hanouneh, H.U. Amin, N.M. Saad and A.S Malik. "The correlation 248, 1995.echnical Writer’s Handbook. Mill Valley, CA: University
between EEG asymmetry and memory performance during semantic Science, 1989.
memory recall," 2016 6th International Conference on Intelligent and [15] D.A. Preece and G.H. Freeman, “Semi-Latin squares and related
Advanced Systems (ICIAS), Kuala Lumpur, 2016, pp. 1-4. doi: designs. J.Roy.Statist.Soc. Ser. B45, 267-277, 1983, retrieved from:
10.1109/ICIAS.2016.7824041. https://doi.org/10.1111/j.2517-6161.1983.tb01250.x.
[10] Mc Hafeezullah Amin, and A.S. Malik “Human memory retention and [16] R.A. Bailey and P.E. Chigbu, Enumeration of Semi-Latin Squares,
recall processes A review of EEG and fMRI studies”, Neurosciences Discrete math, 1997, 167/168, Pg 73-84, Retrieved from:
2013; Vol. 18 (4). https://doi.org/10.1016/S0012-365X(96)00217-8.
[11] M.S. Ahmed and S.A. Yasir “Examining the Effect of Interference on [17] R.A. Bailey. “Efficient Semi-Latin squares”. Statistica Sinica 1992,
Short-term Memory Recall of Arabic Abstract and Concrete Words Vol.2 (413-437).
Using Free, Cued, and Serial Recall Paradigms”, Advances in Language [18] R.N. Mohan, M. Ho Lee, and S.S. Pokhrel, On Orthogonality of Latin
and Literary Studies, December 2015, Vol. 6 No. 6, ISSN: 2203-4714. Squares, J. Comb. Infor. System Sci., 2005, Vol.30(1-4), Pg151- 179.
Retrieved from: https://www.researchgate.net/publication/1959255_
On_Orthogonality_of_Latin_Squares
461 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—When someone walks, there is a repetitive identification also has an important role in video surveillance
movement or coordinated cycle that forms a gait. Gait is and access control systems for supervision and security, for
different, unique and difficult to imitate. This characteristic example in security sensitive environments such as airports,
makes gait one of the biometrics to find out one's identity. Gait banks and certain spaces. Other information can be obtained
analysis is needed in the development of biometric technology, through identification of gait such as age, race, and gender.
such as in the field of security surveillance and the health sector
to monitor gait abnormalities. The center of mass is the unique In order to identify human gait, image processing is
point of every object that has a role in the study of humans needed, which consists of several stages, such as capturing
walking. Each person has a different center of mass. In this gait videos using a camera. Captured videos that consist of a
research, through a series of processes in image processing such set of image frames are extracted so that they can be processed
as video acquisition, segmentation, silhouette formation, and frame by frame to produce a silhouette image. Feature
feature extraction, the center of mass of the human body can be extraction of silhouette images is the key step in identifying
identified using a webcam with the resolution of 640 x 480 pixels gait. Silhouette images represent binary maps of human
and the frame rate of 30 frames/second. The results obtained walking, forming strong features to represent gait because
from this research were gait frames of 510 frames from 17 they capture the movements of most parts of the human body
pedestrian videos. Segmentation process using background [3]. Feature extraction has been carried out in existing studies
subtraction separates the pedestrian object image from the
such as extraction of the entire human body [4] or some limbs
background. Silhouette gait was produced from a series of image
such as the waist, hips, feet [5].
enhancement processes to eliminate noise that interferes the
image quality. Based on the silhouette, feature extraction A number of research related to gait extraction features
provides the center of mass to distinguish each individual's gait. have been carried out. As proposed in [6], gait extraction
The sequence of center of mass can be further processed for based on features of distance and angle between the two legs
characterizing human gait cycle for various purposes. using Hough Transform was proposed. This study produced a
silhouette and skeleton gait. Different approach was shown in
Keywords—Human gait; center of mass; silhouette; feature
[5], where gait was analyzed using DGait database. Feature
extraction; gait cycle; people identification
extraction from 2D and 3D body silhouettes for gait
I. INTRODUCTION identification was performed. Support Vector Machine (SVM)
kernels were used for classification. This research successfully
Walking is a movement that allows one to move from one compared the two features (2D and 3D). In [7], Multi-scale
place to another by moving the foot forward in the correct Principal Component Analysis (MSPCA) was proposed,
position alternately [1]. Repeated movements or coordinated which performed gait recognition based on modelling limbs
cycles form a gait. Every individual's gait is different, which using a spline curve. The feature extraction uses the CASIA-B
makes it unique and difficult to imitate. These characteristics Gait Database silhouette dataset. For classification Neuro-
then make gait one of the biometrics to find out one's identity. Fuzzy and K-Nearest Neighbors (KNN) was used. Another
Biometrics is a technology of recognition and approach was introduced in [8], where Kinect camera sensor
identification based on physiological or behavioral was used for acquisition. The feature extraction process used
characteristics possessed by humans such as gait, face, voice, static features and dynamic features such as wrists, ankles,
iris, fingerprint, etc. [2]. However, biometric recognition with body, knees, shoulders, arms and thighs. K-Nearest Neighbors
face, sound, iris, and fingerprint cannot be done remotely and (KNN) was used for gait classification. The research output is
requires interaction with the subject to be observed. In a database containing 20 pedestrians walking from right to left.
contrary, gait biometric not require direct contact with the It was shown that image frames and silhouettes used in the
subject to be observed making the image acquisition of gait previous research were not captured and processed in real time
can be performed easily in public places as well as captured but based on provided datasets. Feature extraction has been
remotely. Gait is difficult to be hidden and engineered. This done on several members of the human body but has not used
characteristic is very important in the surveillance system. the Center of Mass (CoM) that has a role in the study of
Gait identification is needed in various applications, such humans walking. Therefore, real-time process of acquisition,
as in the health sector where identification is intended to silhouette generation, and feature extraction based on CoM are
identify the type of disease from abnormal gait motion. Gait proposed in this research. The importance and findings of this
462 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
463 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
the morphological process, grayscale images were converted The final stage in silhouette formation is cropping or
into binary images using the thresholding as shown in (5). cutting to produce image frames that focus on gait objects.
This cropping process requires the position of xmin, ymin, width,
As stated in Eq. 3, the dilation operation closed the gap height. The position of xmin was obtained by finding the
between two objects by adding pixels around object A to the minimum value of the column, the position of ymin obtained by
size of the structure of element B. finding the minimum value of the row. The width value was
A B z | B z A 0 (3)
obtained through reducing the maximum column value to the
minimum column value, and the height value obtained by
Erosion operation eroded or reduced the area of the object reducing the maximum row value to the minimum row value.
according to the size of the structure of element B. Table IV shows the silhouette formation process which
A B {z | ( B) z A} consists of four processes. The first column is the sequence of
(4) image processing, and the second column is the result of each
Thresholding process is used to convert the image into enhancement process to improve image quality, namely
binary image as presented in Eq. 5. The threshold ( ) filtering, thresholding, morphology, and cropping.
requirements and desired values were adjusted based on the
TABLE. IV. THE RESULT OF SILHOUETTES FORMATION PROCESS
needs.
1 if f ( x, y ) T Image Processing Result
g ( x, y )
0 if f ( x, y ) T (5)
Grayscale imagery, there is
noise that can be a nuisance and
must be repaired.
--
464 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
D. Feature Extraction
Feature extraction is the process of extracting features 7 85 182
from each silhouette image to get the CoM of each silhouette
image. The CoM (centroid) was generally obtained by using
the average coordinate ( ) value of each pixel composes the
object [10]. The center of mass value was stored in the matrix
in the form of .mat. As shown in (6), CoM of an object was 8 76 177
obtained by calculating the number of pixel in the silhouettes.
n m
N B( x, y )
x 1 y 1 9 63 189
(6)
Eq. (7) and Eq. (8) shows the calculation the average CoM
on the axis of coordinates ( x , y ) based on the number of
pixels [10].
10 51 185
n m
1
x
N
B( x, y)
x 1 y 1
(7)
n m 11 37 177
1
y
N
B( x, y)
x 1 y 1
(8)
The results of the silhouette feature extraction and the
12 39 186
CoM are shown in Table V.
1 89 178
14 52 196
2 93 175
15 57 189
3 97 179
16 64 186
4 99 179
17 71 185
5 98 179
18 81 185
6 94 181 19 86 185
465 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
20 88 186 32 54 185
21 89 184 33 64 177
22 82 186 34 70 175
23 81 180 35 79 171
24 79 182 36 86 175
25 73 191 37 91 181
26 62 208 38 95 180
27 50 185 39 93 185
28 38 172 40 89 180
29 39 171 41 88 171
30 40 174 42 82 164
31 44 184 43 77 164
466 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
44 67 179 56 89 193
45 60 178
57 90 191
46 52 170
58 94 195
47 41 176
59 89 198
48 41 197
60 84 197
53 73 201
54 79 197
55 83 197
467 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
REFERENCES
[1] Carpentier, J., Benallegue, M., and Laumond, J. P., “On The Centre Of
Mass Motion In Human Walking,” International Journal Of Automation
And Computing Vol. 14, Issue 5, October 2017. Pp 542-551. 2017.
[2] Alsaadi, I., M., “Physiological Biometric Authentication Systems,
Advantages, Disadvantages And Future Development: A Review,”
International Journal Of Scientific & Technology Research. Vol. 4, Issue
12, October 2015. Pp 285-289. 2015.
[3] Boulgouris, N., Plataniotis, K., and Hatzinakos, D., “Gait Recogniton
Using Linear Time Normalization,” Pattern Recognition. Vol. 39, Issue
5, May 2006. Pp 969-979. 2006.
[4] Sudha, L.R. and Bhavani, R.,. “Gait Based Gender Identification Using
Statistical Pattern Classifiers,” International Journal of Computer
Applications. Vol. 40, Issue 8, Pp.30-35, 2012.
[5] Borràs, R., Lapedriza, À., and Igual,L.,. “Depth Information In Human
Gait Analysis: An Experimental Study On Gender Recognition”.
International Conference Image Analysis and Recognition. Vol. Part II,
Fig. 5. Movement of the Center of Mass on the Y-Axis. June 2012. Pp 98-105. 2012.
[6] Hustinawaty,. “The Prototype Of Non-Intrusive Skeleton Detection &
III. CONCLUSION Feature Extraction Software & Gait Man In Real Time,” International
Journal Of Sports Science & Engineering. Vol. 07, No. 01, September
In this research, a frame extraction method based on 2012, Pp. 003-022. 2012.
silhouette and center of mass is presented. The results [7] Sai, S. R., Ravi, R., “Multi-Scale Principal Component Analysis Based
obtained from this research are 510 frames that were extracted Gait Recognition,” 1st IEEE/IIAE: International Conference On
from 17 pedestrian videos. Background subtraction process Intelligent Systems And Image Processing 2013 (ICISIP2013).
was successfully separate the gait images from the background. [8] Ahmed, M., “Kinect-Based Human Gait Recognition Using Static and
Dynamic Features,” (IJCSIS) International Journal of Computer Science
The gait silhouette images were acquired after performing a and Information Security”. Vol. 14, No. 12, December 2016.
number of stages by processing grayscale image starting from
[9] Gonzalez, R. and Woods, R., Digital image processing. 4th ed. Pearson,
the noise reduction process to cropping. Based on the 2018.
silhouette image, feature extraction was performed to obtain [10] M.Sayed, “Biometric Gait Recognition Based On Machine Learning
the coordinates of the CoM ( ) for each gait silhouette. The Algorithms,” Journal of Computer Science. Vol.14, Issue 7, July 2018,
results have shown that the CoM in all image frames were Pp 1064-1073, 2018.
successfully identified. For future work, the CoM can be used
as a feature in conducting gait classifications.
468 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Traffic congestion could be a serious road traffic In addition, parking on the road sides can have an impact
problem particularly at intersections because of its potential on traffic flows mainly if it is closer to the intersection as it
impact on the risk of accidents, vehicle delays and exhaust reduces road capacity. Furthermore, side parking can delay or
emissions. In addition, illegal parking by road users at block any movement of vehicles, particularly for left turnings
intersections can give additional deterioration to the intersections [4]. In India, it was reported that parking was one of the
that may create additional traffic flow interruptions. This paper serious problems in urban areas that the country was facing
presented assessment of the illegal parking impact on signalized due to the increment of vehicle ownership and the
intersection at Parit Raja, Malaysia using simulation approach development of mall within the city center [5]. Malaysia is
using PTV VISSIM simulation software. The results showed that
also experiencing similar problems. This is including the
if illegal parkings at Parit Raja intersection were banned, traffic
delay and travel time of vehicles will be improved and thus,
illegal parking in unauthorized areas, particularly near the
improving the intersection Level of Service. road intersection. Illegal parking on roadsides reduces the
effective width of the lane and hence reducing the speed and
Keywords—Traffic simulation; traffic flow; signalized capacity of the prevailing roadways. As a result, the traffic
intersection; level of service; illegal parking will be interrupted which cause delays, accidents, congestion,
etc.
I. INTRODUCTION
In previous years, the assessment of the signalized
Congestion was associated with the necessity and ability to intersections performance had used various approaches that
own personal vehicle. This was interrelated to an increment in focused on the application of software. Simulation by using
population and income level that people become affluent to software was one among the most popular methods including
owning vehicles [1]. The increment in traffic volumes on the SUMO, TRANSIMS and PTV VISSIM [6]. These software
road contribute to the movement conflict, long queue and stop programs were able to produce simulations of traffic flows
delays at intersections. In order to reduce congestion at an and various scenes of traffic operation at signalized
intersection, traffic signal is introduced. Signalized intersections without disturbing the traffic at the actual
intersection allows the traffic to cross the road safely as the locations to detect any problem occurring at the signalized
traffic will be directed to passage in sequence and allow them intersections. The significant of this study was evaluation on
to cross without obstacles. Signalized intersection, apparently, the impact of illegal parking by measuring level of service
can reduce right angle accidents, increase road capacity, bring (LOS) on the road for two scenarios, which are road with
confidence to drivers to cross the roads and provide a good illegal parking and road without illegal parking.
level of service. Although the main function of traffic lights is
to reduce the traffic conflict at intersections, installing traffic II. CASE STUDY: PARIT RAJA SIGNALIZED INTERSECTION
control devices at intersection areas do not always give an This study had been conducted at a signalized intersection
advantage to road users. For example, installing unnecessary located at Parit Raja, Batu Pahat, Malaysia (Fig. 1). The town
traffic control devices can result in disturbing the traffic flows of Parit Raja was one of the busiest areas in Batu Pahat as it is
instead of improving it. The installation of traffic lights at the main destination for businesses and occupations. There
intersections can also create other harms to road users such as were shop lots, banks, offices, industrial areas and educational
rear-end crashes and traffic delays [2]. Rear-end accidents at institutional buildings within the small town. Parit Raja
signalized intersections usually occurred because of the signalized intersection was an important junction which
leading vehicles’ sudden stop due to signal change or traffic connects two main cities, Kluang and Batu Pahat.
situation and drivers’ eyes tend to focus on the traffic light
rather than the vehicle right in front of them [3]. This kind of
situation will slow down the traffic flows at intersections.
469 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
470 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Vehicle
From Batu Pahat From Kluang From Parit Raja Laut From Parit Raja Darat
movement
Movement
Direction / Turn Throu Turn U- Turn Throu Turn U- Turn Throu Turn Turn Throu Turn
Type of Right gh Left Turn Right gh Left Turn Right gh Left Right gh Left
Vehicle
Car 244 1200 180 64 112 904 96 80 156 112 16 212 60 216
Bus 0 8 0 0 0 8 0 0 0 0 0 0 0 0
HGV 8 68 4 0 4 136 8 0 12 0 0 4 4 8
Total 416 1608 264 96 168 1240 212 124 228 284 44 276 216 316
Volume
17.5 67.5 11.1 4.0 9.6 71.1 12.2 7.1 41.0 51.0 7.9 34.2 26.7 39.1
Percent
Total
2384 1744 556 808
Volume
% Car 58.7 74.6 68.2 66.7 66.7 73.2 45.3 64.5 68.4 39.4 36.4 76.8 27.8 68.4
%
39.4 20.7 30.3 33.3 31 15.5 50.9 35.5 26.3 60.6 63.6 21.7 70.4 29.1
Motorcycle
% Bus 0 0.5 0 0 0 0.65 0 0 0 0 0 0 0 0
% HGV 1.9 4.2 1.5 0 2.4 10.6 3.8 0 5.3 0 0 1.5 1.9 2.5
Phase 1 Phase 2
471 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
472 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
V. RESULT AND DISCUSSION Fig. 8 represented as cars that were parked illegally. Parking
From VISSIM simulations, various results could be lot was added in the simulation to represent illegal parking
obtained to show intersection performances such as vehicle areas.
delays and queues. In this analysis, the random seeds used Tables IV and V show the analysis outcome of delay and
were 32, 42 and 52. Time interval for the result was 100 for travel time of the existing traffic (with the presence of illegal
every 600 second. parking). Average delay stop for this simulation was 44.15
Fig. 8 shows 2D graphical result of the existing traffic second and average travel time for the vehicle was 139.62
condition (with the presence of illegal parking) while Fig. 9 second. This showed that, each driver had to wait at least 2 to
shows the condition if illegal parking was eliminated. 3 minutes to cross the intersection.
Vehicles in the simulation for this study had been prepared Table VII shows the queue length results for every random
with different colors to characterize four types of vehicles: seed on each traffic direction. The highest queue length was
(1) red for cars, (2) yellow for motorcycles, (3) green for from Batu Pahat to Kluang with 74.74m length and the
buses, and (4) blue for heavy good vehicles. The white cars in shortest queue length was from Kluang to Parit Raja Laut with
2.6m distance.
Fig. 8. Simulation of Signalized Intersection Condition at Study Location for Existing Condition (with Illegal Parking).
Fig. 9. Simulation of Signalized Intersection Condition at Study Location without Illegal Parking.
473 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Vehicle movement Queue length(m) TABLE. VIII. COMPARISON RESULTS OF STOP DELAY BETWEEN
SIMULATION TRAFFIC FLOW WITH AND WITHOUT ILLEGAL PARKING
To Kluang 24.13
From Parit Raja Without illegal With illegal
To Parit Raja Laut 24.13 Condition
Darat parking parking
To Batu Pahat 47.86
To Kluang 74.74 Random seed (32) 34.80 44.28
From Batu Pahat To Parit Raja Darat 5.07
Random seed (42 32.74 42.03
To Parit Raja Laut 34.59
To Kluang 23.02 Random seed (52) 36.96 46.15
From Parit Raja
To Parit Raja Darat 14.67 Average 34.83 44.15
Laut
To Batu Pahat 14.67
To Parit Raja Laut 2.60 TABLE. IX. COMPARISON RESULTS OF TRAVEL TIME BETWEEN
From Kluang To Parit Raja Darat 20.66 SIMULATION TRAFFIC FLOW WITH AND WITHOUT ILLEGAL PARKING
To Batu Pahat 70.74
Without illegal With illegal
Further analysis, level of service at the Parit Raja Condition
parking parking
intersection could be evaluated by referring to the Highway
Capacity Manual (HCM) [9] level of service on Table VI. Random seed (32) 92.11 134.28
Delay result from the simulation was used as a base to obtain Random seed (42 94.40 149.41
LOS based on HCM. From the simulation results, delay of Random seed (52) 90.83 135.18
44.15 for the studied intersection falls in the category of LOS
D. This showed that the intersection was approaching unstable Average 92.45 139.62
flow (tolerable delay, occasionally wait through more than one
signal cycle before proceeding). TABLE. X. COMPARISONS OF LOS WITH AND WITHOUT PRESENT OF
ILLEGAL PARKING
The impact of illegal parking elimination on Parit Raja
intersection: In Parit Raja, illegal parking has become norm Level of service
Condition Delay result (second)
(second)
for local people which, then created problems with the
surrounding traffic. Business activities on the road curbs by Without illegal parking 34.83 C(>20-35)
the road sides encourage illegal parking in Parit Raja With illegal parking 44.14 D(>35-55)
particularly at the areas near to the intersection.
474 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
ACKNOWLEDGMENT
The authors gratefully acknowledge the contribution of the
Ministry of Education and Universiti Tun Hussien Onn
Malaysia. This paper was prepared as part of the research
report under Research Grant of GPPS Vot H012.
REFERENCES
[1] J,Aderamo,A. Traffic Congestion at Road Intersections in Ilorin,Nigeria.
Mediterranean Journal of Social Sciences, Volume 3(2),(2012),pp. 201-
213.
[2] Washington State Department of Transportation Manual.(2015). Chapter
1300-Intersection Control Type-Design Manual M22-01. In Intersection
Control Type Design Manual (pp.1-20).
[3] Yan, X.,Radwan,E. and Abdel-aty,M. Characteristics of rear-end
accident at signalized intersection using multiple logistic regression
model. Accident Analysis and Prevention, Volume 37,(2005), pp.983-
995.
[4] Morillo Carbonell,C. and Campos Cacheda,J.M. Effect of Illegal On –
Street Parking On Travel Times In Urban Environment. CIT2016-XII
Congreso de Ingeniera del Transporte Valencia,(2016),pp.2491-
Fig. 10. Showing the Traffic Interruption due to Car Exit from Illegal Parking 2503.Available at: http://dx.doi.org/10.4995/CIT2016.2016.3521.
on Road. [5] Boro,D.,M,A,Ahmed.,and Goswami,A.Impact of On Street Parking on
Traffic Flow Characteristics.Available at: https://www.academia.edu/
VI. CONCLUSION 25582493/Impact_of_OnStreet_Parking_on_Traffic_Flow_Characteristi
cs? Auto.
The road network and base data were collected on-site and [6] Tianzi,C.,Shaochen,J.I.N. and Hongxu,Y. Comparative study of
have been used in developing the simulation model using PTV VISSIM and SIDRA on signalized intersection. Procedia-Social and
VISSIM software. This was to ensure that the model could Behavioral Sciences,96(Cictp)(2013),pp.2004-2010.Available at:
replicated the actual traffic condition. Other than that, PTV http://dx.doi.org/10.1016/j.sbspro.2013.08.226.
VISSIM needed to be calibrated and validate to make the data [7] PTV Group,(2004).VISSIM 4.10 User Manual, North America:PTV
reliable [10]. Planung Transport Verkehr AG.
[8] Nurul Nasuha,N.A. and Munzilah,M.D. Overview of Application Of
Based on the simulation results and video recording of the Traffic Simulation Model.MATEC Web Conferences,(2018), Available
study location in this study, illegal parking contributed to at:https:doi.org/10.1051/matecconf/201815003006.
traffic flow disturbances. Removing illegal parking will help [9] Highway Capacity Manual,(2000).Signalized Intersection 4th ed.,
to upgrade the LOS, thus, contributes to smooth traffic flow. United Stated Transportation Research Board of the National Academies
of Science.
This was shown from the improvement of vehicle delay from
[10] Rrecaj,A.A. and M,Bombol,K. Calibration and validation of the VISSIM
44.15 to 34.83 second. Besides that the travel time also has parameter-state of the art. Technology, Education, Management,
been found to improve by 20% which is 47.17 second less Informatics, Volume 4(3),(2015),pp.255-269.
from when the illegal parking existed.
475 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Currently, all the devices are using Internet the IPv4 MPLS backbone current, several ways to connect to
protocol version 4 (IPv4) to access the internet. IP addresses of IPv6 islands. Because the cost of updating the spine in whole
the IPv4 are now depleted from IPv4 pool announced by IANA or in part is greater and needs network updates, therefore
(Internet Assigned Number Authority) in February 2011. To transition mechanisms are deployed. The theory of MPLS is
solve this issue Internet protocol version 6 (IPv6) is launched. But developed and considered as the hybrid technology of ATM
the main problem is current devices can’t support directly IPv6 and IP. This paper evaluates distinct techniques for clouting
that causes various compatibility issues. Many researchers have current IPv4 network MPLS additional IPv6 facilities lacking
proposed various techniques, but still, their efficiency and the need for backbone adjustments. These techniques are used
performance is a big challenge. This study examines several
to isolate IPv6 domains to interact on the present IPv4 MPLS
mechanisms of transition IPv6 the backbone of multiprotocol
label switching (MPLS) to evaluate & compare their
backbone [11]. In the IPV6 tunnels among customer edge and
performances. It involves comparing different performance customer edge CE-to-CE routers together with manual
metrics and manual tunneling tunnel efficiency metrics. The tunnels, ISATAP tunnels, 6to4 tunnels. IPV6 tunnels between
main goal of this paper is to examine the dissimilar tunneling supplier edge and supplier edge PE-to-PE routers along with
techniques and find out which tunneling method is better in all manual tunnels, IPV4 automatic tunnels, ISATAP tunnels, and
performance, which increases network performance. 6to4 tunnels. This paper analyzes performance parameters, i.e.
Experimental results show that ISATAP is better performance in delays in data packets, jitter, and throughput of the network
all metrics. above mentioned techniques and performs statistical analysis
[12-15]. The purpose of doing this study is to investigate the
Keywords—ISATAP; tunneling techniques; IPv4; IPv6; various tunneling mechanisms which run both network IPv4
network performance and IPv6. After the deployment of these mechanisms find out
the best transition mechanism that provides the highest
I. INTRODUCTION throughput with very low delay and jitter in the network. For
Due to the rapid growth of population demand for IP better understanding, the scenarios are shown below in Fig. 1.
addresses has increased more and more [1-3]. Eventually, IP The figure shows this research paper consist of four phases.
addresses of IPv4 pool is completely exhausted. IANA Each phase provides the evaluation of the research intention is
announced on 3 February 2011 that no. of IPv4 addresses are acclimated. The emulation is done by Graphical Network
almost exhausted. Many companies and organizations are Simulator (GNS3) tool. Network simulation and data
moving towards IPv6 addresses. IPv4 is 32 bit long and gathering are done by (OPNET) tool. In the last phase data
supports an address of only 32-bit, meaning 4.3 billion. IPv6 analysis is conducted by MS-office 2013.
is 128 bit long and includes an enormous amount of addresses, A. Contributions/ Findings
i.e. trillions of trillion addresses are now accessible. MPLS is
a mechanism for packet labeling [4-7]. It is extremely scalable The findings of these studies and contributions described
system and commonly utilized in transmission technology by as follows:
internet service suppliers. It plays a vital role in the IPv4 Proposed work covers the shortage of IPv4 addresses
backbone network for companies. MPLS examines the labels and provides full IPv6 connectivity.
and forward data packets discovered on the label instead of
searching for hard routing and examine the packets. Proposed study assessment of a sequence of IPv6 multi-
Companies, use the backbone of MPLS to link offices and protocol label switching (MPLS) transition
sites together remotely. The integration of IPv6 facilities in the mechanisms.
MPLS infrastructure can be seen as ordinary progress by
Analyzing the transition system and identifying which
service suppliers and businesses using MPLS networks [8-
mechanism is best performed in terms of the smallest
10].The MPLS backbone provides the option of connecting
delay lowest jitter and the greatest performance.
IPv6 network, using the existing IPv4 network. When using
476 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Create simulation
III. PROPOSED ANALYSIS
network scenario
For simulation optimized network engineering tools
(OPNET) has been used [16]. The Customer edge to customer
edge (CE to CE) and the provider edge to provider edge (PE to
Data gathering
PE) routers are placed. For configuration of routers well-
known emulation is used, called graphical network simulator
(GNS3) [17-19]. For simulation, all configurations are
Emulation (GNS-3) imported to the OPNET environment. External Border
Gateway Protocol (EBGP) and Multiprotocol Border Gateway
Protocol (MP-BGP) are used for PE routers, CE routers for
remote access to PE router in MPLS whereas Interior Gateway
Simulate of results
(OPNET) Protocol (IGP) and Open Shortest Path First (OSPF) are used
inside the MPLS. For suitable deployments configured for
IPv4 and IPv6 networks, but it also depends on the transition
mechanisms. MPLS cloud set to be in IPv4-enabled and IPv6-
Analysis of results
enabled customers and servers for the transition processes. If
customers need to interact with servers on separate islands on
Fig. 1. Proposed Methodology Flowchart.
each IPv6 island, they must cross the cloud of the IPv4 MPLS.
Then a total of eight tunneling scenarios were configured for
Our research paper is divided into six sections, the second
the various tunneling mechanisms shown in Table II. All
section of the paper describes some existing techniques and
customers and servers have IPv6 allowed configuration. These
drawbacks, the third section presents the methodology of the
tunneling processes were used to traffic IPv6 throughout the
proposed analysis, and the fourth section is about the
current IPv4 network by encapsulating IPv6 packets in the
simulation results and their discussion. The fifth section is all
IPv4 header. Data packet will be decapsulated at the end node
about analysis of data through different statistical methods like
of the tunnel, and it will be removed from the IPv4 packet
ANOVA, F-test and T-test, in the last section is the
header. An actual data packet of IPv6 transferred to Well-
conclusion.
matched Tunnels, and it configured remaining four routers
II. RELATED RESEARCH among provider edge routers.
The various researchers have proposed different tunneling TABLE. II. IPV6 TRANSITION MECHANISMS
mechanisms, but there are some drawback is still in the
network. Following Table I show some tunneling mechanisms Manual tunnel
and their drawbacks. IPv4 MPLS backbone Manual tunnel CE to CE
TABLE. I. SUMMARY OF PREVIOUS WORK AND THEIR DRAWBACKS IPv4 MPLS backbone Manual tunnel PE to PE
Automatic tunnel
S. No. Author Technique Drawbacks
Tunneling, NAT, Old technique and IPv4 MPLS backbone Manual tunnel CE to CE
01 Dr vadym Kaptur
Dual-stack not ideal
A circumstance IPv4 MPLS backbone Manual tunnel PE to PE
Dual-stack and manual
02 Luke smith where point-to- 6to4 tunnel
tunnel
multipoint tunnels
OSPF V3 in IPV6 Only focus on IPv4 MPLS backbone Manual tunnel CE to CE
03 Zeeshan Ashraf
tunneling methods OSPF protocol IPv4 MPLS backbone Manual tunnel PE to PE
Traffic sent from IPV4
This research only GRE tunnel
network to IPV6
focuses on one
network and only
04 M. S. Ali method, but other IPv4 MPLS backbone Manual tunnel CE to CE
method is used that is
methods are
6to4 method in IPv4 MPLS backbone Manual tunnel PE to PE
remaining
OPNET tool
Onlyfocused delay ISATAP tunnel
05 Sami Salih 6vpe performance IPv4 MPLS backbone Manual tunnel CE to CE
parameters.
ISATAP, 6RD and IPv4 MPLS backbone Manual tunnel PE to PE
Dual Stack
Yashwin Sookun, Performance Analysis Network and The CEs were intended to be allowed for IPv4 and IPv6 in
06
Vandana Bassoo of IPv4/IPv6 traffic load is high the event of CE-to CE tunneling, and only IPv4 was configured
Transition for PE routers and all IPv6-enabled client and server
Techniques configuration. After the encapsulating process of IPv6 data
N.
07 Chuangchunsong DS-Lite, 4over6,
Metrics are not packets in the header of the IPv4 network, these tunneling
defined clearly procedures were used to traffic IPv6 across the existing IPv4
at al
Tunneling network. The packet will be decapsulated at the tunnel end
Mohammad
08
Aazam et al
Teredo, ISATAP overhead in node and the packet header IPv4 will be deleted. It will then
Teredo forward the initial IPv6 packet to its final IPv6 place. Next, the
477 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
V. SIMULATION SETUP
The five thousand four hundred values gathered for every
metric are highly greater, as shown in the statistics above the
numbers are close to one another. As a consequence, the
statistical investigation is conducted to define if the
mechanisms for each performance metric have any
statistically significant distinctions. The statistical analysis
conducted to assess the information gathered is described in Fig. 5. OPNET Average Throughput for All Scenarios.
Section 4.
VI. ANALYSIS OF THE RESULTS
OPNET Average End-to-End delay for all scenarios as
shown in Fig. 3. X-axis is total time in minute‟s total running Investigation of the collected data and evaluated the
time for simulation is 5 hours and Y-axis is time for delay resultant data. Methods Below were used to perform the
measured in seconds. statistical analysis.
OPNET Average End-to-End jitter for all scenarios as Analysis of variance (ANOVA)
shown in Fig. 4. X-axis is total time in minute‟s, total running F-TEST
time for simulation is 5 hours and Y-axis is time for jitter that
is delay that is IP delay variation is seconds. T-TEST
OPNET average End-to-End Throughput for overall ANOVA was used to determine if there is statistically
simulation as shown in Fig. 5. X-axis is total time in minute‟s significant difference in means among the scenarios F-Test
total running time for simulation is 5 hours and Y-axis is time was used to determine whether modifications were equivalent
for throughput of the network that process how many packets and whether the mean of one system differed from the mean
can process in a given amount of time. of the other. Finally, either two sample T-test using degree of
freedom was used to determine if the mean of one mechanism
is different from the mean of another mechanism.
478 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A. Scenario. 01 Customer Edge to Customer Edge (CE-to- TABLE. IV. F-TEST RESULT FOR DELAY CE-TO-CE TUNNEL
CE) F-TEST n1,n2 is the degree of freedom f0=S12/S22
To calculate significant variations between the metrics of Transition F α, n1- Test if F0>F α
F0
IPv6 CE-to-CE tunneling processes (delay, jitter, throughput) Mechanisms 1, n2-2 n1-1,n2-2
and Calculate which method is the best. Auto CE & 6to4 CE one 0.961206111 Negative
1) Analysis of delay: For ANOVA, the hypothesis below Auto CE & GRE CE one 0.941160555 Negative
has been recognized. • Null Hypothesis (H0): delay implies Auto CE & ISATAP
one 0.801237 Negative
CE
equivalent to CE-to-CE tunneling processes Alternative
Manual CE & 6to4
Hypothesis (H1): for CE-to-CE tunneling processes, at least one 0.963112573 Negative
CE
one delay implies different from other means. Manual CE & GRE
one 0.943143 Negative
CE
TABLE. III. ANOVA RESULTS FOR DELAY CE-TO-CE TUNNEL Manual CE &
one 0.85365 Negative
ISATAP CE
ANOVA: Single Factor 6to4 CE & GRE CE one 0.979155778 Negative
Specified group Total Sum (Quantity) Average Difference 6to4 CE & ISATAP
Manual CE- one 0.89125 Negative
51699 107.4725336 0.0020970 2.44585 CE
CE
Auto CE 51762 106.4065222 0.0020822 2.44083 TABLE. V. T-TEST RESULT FOR DELAY CE-TO-CE TUNNEL
GRE CE-CE 51758 109.4823187 0.0031417 2.5934
6to4 CE-CE 51792 108.815763 0.0034448 2.53934 Two Sample T-Test
ISATAPCE- tα Test if t0<-tα ,
51750 105.403224 0.0030568 2.33172 t0
CE Transition mechanism n1+n2-2 n1+n2-2
(contd.)
P- Auto CE & 6to4 CE 1.63 -4.25017473 YES
Source of
SS DF MS F VALU F-critical
variation Auto CE & GRE CE 1.63 -6.06411939 YES
E
Between 0.00012 4.248 2.37 Auto CE & ISATAP -
4 1.63 YES
Groups 7451 35 F test > F CE 3.638063334
critical Manual CE & 6to4 -
Within 0.51942 25921 2.504 16.9600 1.63 YES
5.2 Reject CE 5.453454824
Groups 6123 1 91 9107 Manual CE & GRE -
null 1.63 YES
0.51955 25921 Hypothesi CE 1.820066369
Total Manual CE & -
3573 5 s 1.63 YES
ISATAP CE 2.639061371
The result for delay performance metrics from CE-to-CE -
6to4 CE & GRE CE 1.63 YES
3.999067322
tunnel is shown in above Table III. where F test>F critical for 6to4 CE & ISATAP -
Seeing that null hypothesis is rejected that reason it is enough CE
1.63
2.638067771
YES
evidence that at least one delay mean is different from other
delay mean for all scenarios. 4) Analysis of jitter: Jitter is also analyzed using similar
techniques. The ANOVA results in Table VI were obtained.
2) F-Test for delay CE-to-CE tunnel: F test was used to
evaluate if the variances are equal using the hypothesis given TABLE. VI. ANOVA RSULTS FOR JITTER CE-TO-CE TUNNEL
below.
ANOVA: Single Factor
Null hypothesis (H0)= delay variance i= delay variance Groups Count Sum Average Variance
Alternative hypothesis (H1) = delay variance i=<delay Manual CE-CE 51753 82.6607 0.001597 4.05
variance j. Auto CE-CE 51864 82.52443 0.001591 4.14
GRE CE-CE 51858 85.79141 0.001654 4.28
The result shown in Table IV since F0>F α n1-1, n2-2 is 6to4 CE-CE 51892 84.85763 0.001635 4.37
not true then null hypothesis is not rejected and it is enough
ISATAPCE-CE 51850 81.33167 0.001568 4.28
evidence that delay variances are equal in that condition we
(contd.)
will perform T-test to find which delay mean is less than the
Source
others. of SS Df MS F
p-
F-critical
value
variation
3) T-Test for delay CE-to-CE tunnel: The results are Between
shown in Table V, since t0<-tα, n1-1, n2-2 then the null Groups 0.000143 3 4.77 2.604952
hypothesis was rejected and hence it is enough evidence that SSB F test > F
Within critical
delay means of Manual CE-to-CE and Automatic CE-to-CE is Groups 0.087257 207363 4.21
113.4597 2.8
Reject Null
less than the 6to4 CE-to-CE and GRE CE-to-CE. Additionally SSW
ISATAP CE-to-CE tunnel is less than the GRE CE-to-CE Total
0.0874 207366
Hypothesis
tunnel. SST
479 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
5) F-Test for jitter CE-to-CE tunnel: The results shown in TABLE. IX. ANOVA RSULTS FOR DELAY PE-TO-PE TUNNEL
Table VII says that there is enough evidence to support that Summary
end-to-end jitter means of different tunneling Manual CE-to-
CE and Automatic CE-to-CE are less than 6to4 CE-to-CE and Groups Count Sum Average Variance
ISATAP CE-to-CE tunnel, and 6to4 CE-to-CE has lower Auto PE 52022 107.5243 0.001591 2.45
mean jitter than ISATAP CE-to-CE.
6) T-Test for jitter CE-to-CE tunnel: The results shown in Manual PE 52022 107.5243 0.001597 2.45
Table VIII says that there is enough evidence to support that GRE PE 51938 108.4176 0.001654 2.39
end-to-end jitter means of different tunneling Manual CE-to-
6to4 PE 51938 108.4080 0.001635 2.39
CE and Automatic CE-to-CE are less than 6to4 CE-to-CE and
ISATAP CE-to-CE tunnel, and 6to4 CE-to-CE has lower ISATAP PE 52010 106.5133 0.001568 2.57
mean jitter than ISATAP CE-to-CE. (contd.)
B. Scenario. 02 Provider Edge to Provider Edge (PE-to-PE) Source
of p- F-
1) ANOVA results for delay PE-to-PE tunnel: The result SS df MS F
variati value critical
for delay PE-to-PE performance parameters are shown in on
Table IX. Since Ftest >Fcritical therefore the Null Hypothesis Betwe
2.60495
is rejected and there is enough evidence to demonstrate that at en 0.000127 4.24835
3 2
least one delay mean is different from other delay means Group 451 E-05
F test >
s SSB
among the all scenarios. 16.96009 0.0296
F
Within critical
0.519426 2073 2.50491 107 44
TABLE. VII. F-TEST RSULTS FOR JITTER CE-TO-CE TUNNEL Group Reject
123 63 E-06 Null
s SSW
F-TEST n1,n2 is the degree of freedom f0=S12/S22 Hypothe
Total 0.519553 2073
sis
Transition F α, n1-1, Test if F0>F α n1- SST 573 66
F0
Mechanisms n2-2 1,n2-2
Auto CE & 6to4 CE one 0.945501655 Negative 2) F-Test results for delay PE-to-PE tunnel: The result is
shown in Table X, since F0>F α n1-1, n2-2 is not true then
Auto CE & GRE CE one 0.967344952 Negative
null hypothesis is not rejected and it is enough evidence that
Auto CE & ISATAP
CE
one 0.922723464 Negative delay variances are equal in that condition we will perform T-
Manual CE & 6to4 test to find which delay mean is less than the others.
one 0.945787554 Negative
CE 3) T-Test results for delay PE-to-PE tunnel: The result is
Manual CE & GRE
one 0.976769838 Negative
shown in Table XI, since t0<-tα, n1-1, n2-2 then the Null
CE Hypothesis was rejected. Therefore there is enough evidence
Manual CE &
ISATAP CE
one 0.944407759 Negative to support that end-to-end delay mean of Manual PE-to-PE
and Automatic PE-to-PE are less than the 6to4 PE-to-PE and
6to4 CE & GRE CE one 0.956819337 Negative
GRE PE-to-PE. Additionally it demonstrates that ISATAP PE-
6to4 CE & ISATAP
CE
one 0.945066522 Negative to-PE tunnel is lower delay mean as compare to GRE PE-to-
PE tunnel.
TABLE. VIII. T-TEST RSULTS FOR JITTER CE-TO-CE TUNNEL
TABLE. X. F-TEST RSULTS FOR DELAY PE-TO-PE TUNNEL
Two Sample T-Test
F-TEST
tα Test if t0<-tα ,
Transition mechanism t0 Transition Fα n1-1 Test if F0 > Fα n1-
n1+n2-2 n1+n2-2 F0
mechanism n2-1 1 n2-1
Auto CE & 6to4 CE 1.63 -10.88907171 YES
Auto PE & 6to4 PE one 1.024523462 YES
Auto CE & GRE CE 1.63 -15.68834028 YES
Auto CE & ISATAP
1.63 -11.22907883 YES Auto PE & GRE PE one 1.024800865 YES
CE
Manual CE & 6to4
1.63 -9.441690775 YES Auto PE & ISATAP
CE one 1.02525635 YES
PE
Manual CE & GRE
1.63 -14.25657544 YES Manual PE & GRE
CE one 1.024523462 YES
PE
Manual CE &
1.63 -8.88922133 YES
ISATAP CE Manual PE &
one 1.024800865 YES
6to4 CE & GRE CE 1.63 -4.672126496 YES ISATAP
6to4 CE & ISATAP Manual PE & 6to4
1.63 -9.889071761 YES one 1.024785256 YES
CE PE
480 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
481 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. XVI. BEST TO WORST OVERALL IPV6 TRANSITION MECHANISM [6] D. R. Al-Ani, A. R. Al-Ani, “The Performance of IPv4 and IPv6 in
Terms of Routing Protocols using GNS 3 Simulator”, Procedia
Overall (including delay, jitter, and Computer Science, Vol 130, 2018, Pages 1051-1056.
Ordinal Overall
throughput) IPv6 Transition
Ranking Value Ranking [7] J. Beeharry and B. Nowbutsing, "Forecasting IPv4 exhaustion and IPv6
Mechanisms in Order of Best to Worst
migration", 2016 IEEE International Conference on Emerging
ISATAP PE Delay (0.00204793) jitter
1 1 Technologies and Innovative Business Practices for the Transformation
(0.001603592) Throughput (2632852)
of Societies (EmergiTech), Balaclava, 2016, pp. 336-340.
Manual PE-to-PE and Automatic PE-to-
PE
13 6 [8] Y. Sookun,and V. Bassoo, “Performance analysis of IPv4/IPv6
Manual CE-to-CE, Automatic CE-to-CE, transition techniques”,2016 IEEE International Conference on Emerging
14 7 Technologies and Innovative Business Practices for the Transformation
6to4 PE-to-PE, and GRE PE-to-PE
of Societies (EmergiTech), Balaclava, year 2016 pp. 188-193.
6to4 CE-to-CE , ISATAP CE 21 10
[9] R. Z. Khan, and A. Shiranzaei, "IPv6 security tools—A systematic
GRE CE-to-CE 23 11 review", 2016 International Conference on Computing, Communication
and Automation (ICCCA), Noida, 2016, pp. 459-464. The Illustrated.
VII. CONCLUSION [10] Network How TCP/IP Works in a Modern Network Book2nd Edition
This paper has two phases of contribution, i.e., year 2017 by Walter Goralski.
connectivity of IPv4 and IPv6; secondly, test performance of [11] V. Kher, A. Arman and D. S. Saini, "Hybrid evolutionary MPLS
Tunneling Algorithm based on high priority bits," Futuristic Trends on
different tunneling techniques. From the above simulation test Computational Analysis and Knowledge Management (ABLAZE), 2015
result, ISATAP PE is best because of the high throughput and International Conference on, Noida, 2015.
lowest jitter during data packets transmission. Whereas GRE [12] The Illustrated Network How TCP/IP Works in a Modern Network
CE is worst due to its high jitter and lowest throughput in the Book2nd Edition year 2017 by Walter Goralski.
network. The main objective is to provide IPv6 connectively [13] Hamarsheh, and Ala,“Deploying IPv4-only Connectivity across Local
and test which tunneling technique is better to have better IPv6-only Access Networks”, IETE Technical Review Taylor & Francis
performance than others. Future work can extend in payload year 2018, pp. 1-14.
of the network. Additionally security of these tunneling [14] Technologies and Innovative Business Practices for the Transformation
techniques can be analyzed. of Societies (EmergiTech), Balaclava, 2016, pp. 336-340.
[15] Y. Sookun,and V. Bassoo, “Performance analysis of IPv4/IPv6
REFERENCES transition techniques”,2016 IEEE International Conference on Emerging
[1] R. Tadayoni,and A. Henten, „From IPv4 to IPv6: Lost in translation?‟, In Technologies and Innovative Business Practices for the Transformation
Telematics and Informatics, vol. 33, the year 2016, Issue 2, pp 650-659. of Societies (EmergiTech), Balaclava, year 2016 pp. 188-193.
[2] L. Smith, M. Jacobi and S. Al-Khayatt, "Evaluation of IPv6 transition [16] R. Z. Khan, and A. Shiranzaei, "IPv6 security tools—A systematic
mechanisms using QoS service policies", 2018 11th International review", 2016 International Conference on Computing, Communication
Symposium on Communication Systems, Networks & Digital Signal and Automation (ICCCA), Noida, 2016, pp. 459-464.
Processing (CSNDSP), Budapest, Hungary, 2018, pp. 1-5. [17] The Illustrated Network How TCP/IP Works in a Modern Network
[3] M. Nikkhah, “Maintaining the progress of IPv6 adoption”, In Computer Book2nd Edition year 2017 by Walter Goralski.
Networks, vol 102, 2016, pp 50-69. [18] Hamarsheh, and Ala,“Deploying IPv4-only Connectivity across Local
[4] N. Zhang, M. A. Sirbu and J. M. Peha, "A comparison of migration and IPv6-only Access Networks”, IETE Technical Review Taylor & Francis
multihoming support in IPv6 and XIA", 2017 International Symposium year 2018, pp. 1-14.
on Networks, Computers, and Communications (ISNCC), Marrakech, [19] F. Siddika, and M. A. Hossen , and S. Saha, “Transition from IPv4 to
2017, pp. 1-8. IPv6 in Bangladesh: The competent and enhanced way to follow”, 2017
[5] Kamaldeep, M. Malik and M. Dutta, "Implementation of single-packet International Conference on Networking, Systems and Security(NSysS),
hybrid IP traceback for IPv4 and IPv6 networks", IET Information Dhaka, year 2017, pp. 174-179.
Security, vol. 12, no. 1,year 2018, pp. 1-6.
482 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Smartphone becomes one of the most popular According to a recent study, Google play published more
devices in last few years due to the integration of powerful than 3.5 million apps from 2009 to December, 2017 [5]. The
technologies in it. Now-a-days a smartphone can provide number of apps is rapidly increasing over recent few years.
different services as like as a computer provides. Smartphone Another recent security study showed that, in Google play
holds our important personal information such as photos and store, more than 200 malevolent apps were found [6]. These
videos, SMS, email, contact list, social media accounts etc. apps collected private information like contact numbers, places
Therefore, the number of security and privacy related threats are etc. from users and sent to the attackers‟ server. Time to time
also increasing relatively. Our research aims at evaluating how this information was resending to the attackers‟ server when
much the smartphone users are aware about their security and
users use these apps. In the early 2016, Google banned 13 apps
privacy. In this study, firstly we have taken a survey for
smartphone users to access the level of smartphone security
from Google play store because, these apps collected
awareness displayed by the public. We also determine whether a information from users and sell to other server [7].
general level of security complacency exists among smartphone In this paper, we discuss about results of a security and
users and measure the awareness of android users regarding privacy awareness survey for the smartphone users. The
their privacy. From survey result we have found that, most of the research aims at evaluating how much the smartphone users are
people are not aware about their smartphone security and aware about their security and privacy. In this survey, we
privacy. Secondly, based on survey results, we have shown a
create questionnaire to access the level of smartphone security
method to measure the level of awareness (LOA) for the
awareness displayed by the public. We determine whether a
smartphone users. By using this method, a user can easily
measure his/her smartphone security and privacy related level of general level of security complacency exists amongst
awareness. smartphone users and based on these result we show a statics
model to measure the awareness of android users regarding
Keywords—Smartphone; Smartphone Problems; Level of their privacy.
Awareness (LoA); Security and Privacy This paper is organized as follows. We start with a
I. INTRODUCTION discussion of the various previous related works in Section II.
Then we explain about the smartphone problems in Section III
The technologies of smartphone have been increasing with and discuss different types of attacks in smartphone. In
a huge rate over last few years. Smartphone provides many Section IV, we focus on our research methodology along with
services as data sharing, phone calls, internet, different online pilot study, research instrument and target population, and data
& offline games etc. Therefore, it increases the chance of analysis. In Section V, we analyze the result of our survey
security and privacy related threats comparatively. Almost including research questions, evolution of research question
80% of activities related to the internet, so it is important for us and then propose a model that can measure the level of
to become aware about security and privacy. Several recent awareness. Finally, we show some concluding remarks and
studies shown that, when security comes to smartphone, most future direction in Section VI.
of the smartphone users are propitious [1, 2, 3]. In order to
authentication of smartphone, people often use different II. PREVIOUS WORK
patterns, finger print password, face password, pin passwords Benenson et al. [8] pointed that IT security plays an
etc. All these are not enough to protect us from security related important role while someone use smartphone, because of its‟
issues [4]. Smartphones are handhold device where different broadly acknowledged and well documented feature, which
personal information are stored. We have to ensure the security mainly focused on the technical area of a smartphone security
of our personal information. Most of the time, due to lack of system. According to their interview of 24 users on IT security
our awareness we fail to protect our personal information. If all of smartphone, they found the role of user. Based on this result
this information falls into a bad hand, we might be in trouble. they consecrated five hypotheses and proposed a mental
technique after evaluation of these hypotheses.
483 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A recent study in South Africa by Ophoff & Robinson [9] A. Data Protection and Privacy
shown that the level of awareness on smartphone security Muslukhov et al. [13] found out the problem of data
based on public users and determined how much a common protection and privacy and discussed the types of data a user
security level exists in smartpnone users. According to their wants to protect in smartphone. Authors‟ also showed for the
survey on smartphone security awareness, they examined 619 different types of data how the required security protection is
South African smartphone users based on the trust of change. In another recent study, Muslukhov [14] discussed
smartphone apps and other third party apps. They found that about data protection and privacy problem and showed that the
users showing high level of trust on smartphone apps, rather regular update of smartphone lock screen for users‟
than when they install other third party apps. In this study, they authentication and accessibility creates the security and
used an updated version of model developed by Mylonas et al. protection level more strong.
[10].
B. Attacks
Alani [11] noted that android smartphone privacy
awareness concern grow with spread in users‟ perspective. A Attacks are similar in all smart devices such as smartphone,
huge number of apps are downloaded daily by the users, but it laptop, tablet etc. Attacks in smartphone categorized into two
is really difficult to differentiate between good terms of service parts including: old attack and new attacks. Old attacks include
security apps and bad terms of service security apps. In this physical attacks, different type of smartphone virus, backdoor,
paper, authors shown a result based on a survey of 4027 threats, Trojan, different types of malware, worms, radio and
android smartphone users for android user security awareness. wireless network attacks, and spam attacks. New attacks
According to their survey, they tried to show the interactions include relay attack, counter attack, DOS attack, brute force
between users and terms and service security while they install attack, camera based attacks, SMS based attack, XSS attack,
apps. control-flow attack, etc.
The technologies of smartphone have been increasing with Among all these categories of smartphone problems
a huge rate over last few years. Now-a-days a smartphone can „Attacks‟ is the most common one. Two types of attacks are
provide different services as like as a computer can provide. old attack and new attack. Both attacks have some individual
Our smartphone holds much information such as mailing impact to the smartphone. Table I and Table II show the impact
information, messaging information, calling information etc., of smartphone due to old attack and new attack.
which are very important for us. Therefore, we have to ensure
the security and privacy of our smartphone.
The addition of powerful OSs, applications, hardware etc.,
makes smartphone strong and secure, but all these are not
enough to protect our privacy. As the number of privacy and
security related threats are raising comparatively. The security
and privacy related challenges in smartphone are slightly same
as the computer threats environment. Smartphone problems are
categorized into four parts [12] including: Data protection and
privacy, Attacks, Authorization, and Vulnerabilities (Fig. 1). Fig. 1. Categorization of Smartphone Problems.
484 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. OLD ATTACKS AND THEIR IMPACT TO THE SMARTPHONES phones contain smartphone functionalities, they just use their
Attack Name Impact to the Smartphone
phone for call or SMS related work only. Some user installed
different third party apps without knowing the terms and
Makes the security of smartphone weak
service related conditions. Some users utilize the full
Physical Attack Causes abnormal behavior in smartphone
[15] Unauthorized code can be effect to the users
smartphone functionalities.
privacy B. Research Instruments and Target Population
Causes abnormal behavior in application and
Smartphone An online tool was used here based on the questions to
smartphone
Virus [16,17]
Private information can be leaked via applications analyze the collected data. This research contains 20 questions
Makes the security of smartphone weak and the answers might be one or multiple. All these questions
Backdoor [18]
Create a backdoor for smartphone viruses are based on security and awareness of smartphones. Among
Makes the security of smartphone weak these questions we have used just 7 in our study, which can
Threat [19] Data may be hacked fulfill our goal and objectives. Our aim is to evaluating whether
Creates backdoor into private information
smartphone users aware about their security and privacy
Interfere in smartphone operations related issue, and to evaluate how much aware they are. The
Malware [20,21]
Collects private information
target population of this study was smartphone users,
Data may be hacked especially university students of different countries on the age
Wireless Attack
Makes the security of smartphone weak
[22] group between 20 to 26 ages. The purpose of this study is to
Private information may be leaked.
understand the security and privacy awareness from the
Fill the e-mail inbox with unnecessary information
Decrease the smartphone internet speed
smartphone users.
Spam [23]
Collect different important information like C. Data Analysis and Discussion
contact list, message etc.
At first, we set our questionnaires in a Google form. By
TABLE. II. NEW ATTACKS AND THEIR IMPACT TO THE SMARTPHONES using this Google form, we have taken survey from university
students‟ age group in between 20 to 26 year. Then we have
Attack Name Impact to the Smartphone stored these results in Microsoft Excel format for further use.
Counter Attack After completion the survey, we have found how many
Target information can be accessed
[24]
responses are there, whether everything is okay or not. We also
Relay Attack [25] Private information may be hacked.
check every necessary question is answered clearly or not,
Slow the network
DOS attack [26] whether the result fulfills our objectives. Then we combined
Busy the smartphone services
Camera based Makes the security of smartphone weak
our survey result together and found out our objectives. Since,
attack [27] Collects users private information our problem statement is related to the security and privacy
SMS based attack Slow the smartphone operations awareness of smartphone and we combine the survey results
[28] Collects sensitive information and try to find the level of smartphone security awareness
Collect different important information like displayed by public, whether the general level of security exists
Control flow
attack [29]
contact list, message etc. amongst smartphone users etc. To present our survey results,
Memory information can be accessed we use bar chart. In this study, we have used Google form,
Brute force attack Slow the CPU speed computer, Microsoft Excel to find out the security and privacy
[30] Users password may be hacked awareness of smartphone.
IV. METHODOLOGY V. SURVEY RESULTS
The aims of this research at evaluating how much the In total 3,424 responses recorded in this survey, among
smartphone users are aware about their security and privacy. them 175 (5.11%) responses were rejected during initial
Data collection based on industrial survey is the most common exploration of data analysis because, all required questions
process for research project, but this process requires large time were not answered. Of the remaining 3,249 responses are used
to complete, and data analysis is costly [31]. However, a recent in this study. We have analyzed the survey results based on
study by Couper [32] discussed about the different seven research questions which have discussed in this section.
technologies of data collection, which can be used to analyze All these questions are important to find out the awareness of
the data automatically (e.g. Google form). Another study by smartphone security and privacy because all these questions
Granello et al., [33] pointed that online data collection has are addressed to smartphone problems.
become very popular strategy in many research methodologies.
A. Research Questions
A. Pilot Study
The aim of this research is to measure the level of
In our study, we have used survey strategy to find the smartphone security awareness displayed by the public. Also to
quantitative results. The survey was planned to find out the determine whether, a general level of security complacency
level of security and privacy awareness among the smartphone exists amongst the smartphone users and to measure the
users. To understand the topic better on “security and privacy awareness of android users regarding their privacy. The
awareness survey for smartphone users” we consulted with research questions are planned in very simple language, which
many smartphone users and discussed about their smartphone is easy to understand. All these objectives lead to the following
security related problems. We found three types of users. Some questions:
users treated their phones as normal phone, although their
485 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Q1: For what purpose do you use Smartphone? Q3: Third party applications are not same as the operating
system or manufacture of smartphone, as they are created by
Q2: From where you mostly install applications? vendor. Third party apps contain most of the malware rather
Q3: Do you ever install third party applications or than system apps, that‟s why third party apps are more insecure
applications from Unknown sources in your than system apps. In another scenario, third party apps from
Smartphone? unknown sources are more insecure than third party apps from
built-in source for system (e.g. play store). In our survey result
Q4: Before installing application do you read for question Q3 in Fig. 4, we can see 60% people installed
application provider‟s privacy and policy for using third party apps from unknown sources.
application‟s?
Q4: Before installing apps, the application provider
Q5: Before installing application do you ever read provides the privacy and policy of their apps. This privacy and
through application‟s phone access permissions? policy contains about the policy of information about users‟
access. For example, your application extracts the contact list
Q6: What authentication system do you use to lock information from user, so you must have to notify the user
screen for security? about it. From the privacy and policy user can know where, for
B. Evaluation of Research Questions what and how long his/her information will be used. This is
very important and users should read these privacy and policy
Q1: Now-a-days smartphone can perform different services
before installing application. In Fig. 5, we have shown the
as like computer such as email, SMS, location tracking, contact
survey result for Q4, we can easily observe that 25.50% people
list, stores photos and videos, social media account etc. Q1 is
never read privacy and policy and 52.70% of the people read
about the purpose of using smartphone to find how many
privacy and policy sometimes.
people use all these services in their smartphone. Fig. 2 shows
the result of this question, we can see that only 7.3% of the
people use smartphone just for communication and they are Chart for Q2
less insecure than 88.20% of the people who use smartphone 70.00%
for all these activities.
60.00%
Q2: Since a smartphone provides different facilities such as
50.00%
email, Google drive, SMS, and different social media, etc. It
contains a lot of personal information that is very important for 40.00%
us and we should keep these secure. But most of the time we 30.00%
keep our personal accounts (e.g. Email, Facebook, Google
20.00%
drive etc.) logged in to our smartphone. Suppose, someone lost
his/her smartphone and if personal accounts logged in to the 10.00%
smartphone, he/she might be lost his/her personal information. 0.00%
Since, our study is about security and privacy awareness we Q2
have used this question to find out how much people aware
about their security and privacy. Fig. 3 shows the result of this No Yes
question, and we can see 65.5% people are not aware in this
concern. Fig. 3. Do you Sign out from your Personal Accounts (e.g. Email, Facebook,
Google Drive Etc.) after using it with Smartphone?
Chart for Q1
100.00% Chart for Q3
80.00% 70.00%
60.00%
60.00%
50.00%
40.00% 40.00%
30.00%
20.00%
20.00%
0.00% 10.00%
Q1
0.00%
All of them Browsing Social Websites Q3
Fig. 2. For What Purpose do you use Smartphone? Fig. 4. Do you ever Install Third Party Applications or Applications from
unknown Sources in your Smartphone?
486 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
30.00%
Fig. 5. Before Installing Application, do you Read Application Provider‟s
Privacy and Policy for using Application‟s? 20.00%
C. Proposed Model Fig. 7. What Authentication System do you use to Lock Screen for Security?
Smartphone security is not limited to those six questions
but, when we think about smartphone security and privacy TABLE. III. CONSIDERED OPTIONS FOR QUESTIONS
awareness survey those questions gets the top priority. In
Question Safe option Unsafe option
recent days, those reasons are more responsible for losing the
All
privacy of smartphone. Based on survey result we have Q1 Communication Browsing social websites
developed (1) which can measure the level of awareness (LoA) Web surfing
for a smartphone user. Q2 Yes No
Q3 No Yes
( ) (1)
Never
Q4 Always
Sometimes
Where, Q = Q1+Q2+Q3+Q4+Q5+Q6
Never
Q5 Always
We have considered some safe option and unsafe options Sometimes
for each question which denote to secure and insecure zone Pin Code
Password
respectively. Safe options for every question carry the value of Q6 Nothing
Pattern
0 (zero) and unsafe options carry value of 1 (one). Table III Fingerprint
shows the safe and unsafe options for each question.
487 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. IV. PERCENTAGE LOA FOR THE VALUE OF „Q‟ [12] Zaidi, S. F. A., Shah, M. A., Kamran, M., Javaid, Q., & Zhang, S.
(2016). A Survey on security for smartphone device. IJACSA)
Value of ‘Q’ Percentage LoA International Journal of Advanced Computer Science and Applications,
0 100% 7, 206-219.
1 53.78% [13] Muslukhov, I., Boshmaf, Y., Kuo, C., Lester, J., & Beznosov, K. (2012,
2 23.84% April). Understanding users' requirements for data protection in
smartphones. In Data Engineering Workshops (ICDEW), 2012 IEEE
3 9.49% 28th International Conference on (pp. 228-235). IEEE.
4 3.60% [14] Muslukhov, I. (2012). Survey: Data protection in smartphones against
5 1.34% physical threats. Term Project Papers on Mobile Security. University of
6 0.49% British Columbia.
[15] Kataria, A., Anjali, T., & Venkat, R. (2014, February). Quantifying
smartphone vulnerabilities. In Signal Processing and Integrated
VI. CONCLUSION Networks (SPIN), 2014 International Conference on (pp. 645-649).
This research aims at evaluating how much the smartphone IEEE.
users are aware about their security and privacy. In this study, [16] La Polla, M., Martinelli, F., & Sgandurra, D. (2013). A survey on
firstly we have taken a survey from smartphone users to access security for mobile devices. IEEE communications surveys & tutorials,
15(1), 446-471.
the level of smartphone security awareness. We have found
that on average 60% people do not aware about their [17] Cheng, J., Wong, S. H., Yang, H., & Lu, S. (2007, June). Smartsiren:
virus detection and alert for smartphones. In Proceedings of the 5th
smartphone security and privacy. Secondly, we have proposed international conference on Mobile systems, applications and services
a model to measure the level of awareness for smartphone (pp. 258-271). ACM.
users. We have found that almost 50% of the smartphone user [18] Durairaj, M., & Manimaran, A. (2015). A study on security issues in
contains 9.49% level of awareness. Although, the addition of cloud based e-learning. Indian Journal of Science and Technology, 8(8),
new technologies makes a smartphone smarter, the security and 757-765.
privacy related threats also increases relatively. In future work, [19] Pfleeger, C. P., & Pfleeger, S. L. (2002). Security in computing. Prentice
we will extend this study by adding others security and privacy Hall Professional Technical Reference.
related behavior and make our model more efficient and [20] Khouzani, M. H. R., Sarkar, S., & Altman, E. (2012). Maximum damage
malware attack in mobile wireless networks. IEEE/ACM Transactions
accurate. on Networking, 20(5), 1347-1360.
REFERENCES [21] Peng, S. C. (2013). A survey on malware containment models in
[1] Roesner, F., Kohno, T., & Molnar, D. (2014). Security and privacy for smartphones. In Applied Mechanics and Materials (Vol. 263, pp. 3005-
augmented reality systems. Communications of the ACM, 57(4), 88-96. 3011). Trans Tech Publications.
[2] Chin, E., Felt, A. P., Sekar, V., & Wagner, D. (2012, July). Measuring [22] Mandke, K., Nam, H., Yerramneni, L., Zuniga, C., & Rappaport, T.
user confidence in smartphone security and privacy. In Proceedings of (2003). The evolution of ultra wide band radio for wireless personal area
the Eighth Symposium on Usable Privacy and Security (p. 1). ACM. networks. Spectrum, 3, 10-6.
[3] Jones, B. H., & Heinrichs, L. R. (2012). Do business students practice [23] Xu, Z., & Zhu, S. (2012, August). Abusing Notification Services on
smartphone security?. Journal of Computer Information Systems, 53(2), Smartphones for Phishing and Spamming. In WOOT (pp. 1-11).
22-30. [24] Lee, H. T., Kim, D., Park, M., & Cho, S. J. (2016). Protecting data on
[4] Yildirim, N., Daş, R., & Varol, A. (2014, May). A Research on Software android platform against privilege escalation attack. International
Security Vulnerabilities of New Generation Smart Mobile Phones. In Journal of Computer Mathematics, 93(2), 401-414.
2nd International Symposium on Digital Forensics and Security (pp. 6- [25] Yalcin, S. B. O. (2010). Radio Frequency Identification. Security and
16). Privacy Issues. In 6th international workshop, RFIDSec (pp. 8-9).
[5] PhoneArena, "Android's Google Play beats App Store with over 1 [26] Dondyk, E., & Zou, C. C. (2013, January). Denial of convenience attack
million apps, now officially largest," [Online]. Available: to smartphones using a fake Wi-Fi access point. In Consumer
http://www.phonearena.com/news/ [Accessed: 07 July,2019]. Communications and Networking Conference (CCNC), 2013 IEEE (pp.
[6] Dr.Web, "Android.Spy.277.origin," [Online]. Available: 164-170). IEEE.
http://vms.drweb. [Accessed: 07 July, 2019]. [27] Amravati, M. E. S. (2015). A Review on Camera Based Attacks on
[7] Dan, G., "Malicious apps in Google Play made unauthorized downloads, Andriod Smart Phones. International Journal of Computer Science &
sought root,"[Online]. Available: http://arstechnica.com/information- Technology, 6(1), 88-92.
technology/2016/01/malicious-apps-in-google-play-made-unauthorized- [28] Stites, D., & Tadimla, A. A Survey Of Mobile Device Security: Threats,
downloads-sought-root/. [Accessed: 07 July,2019]. Vulnerabilities and Defenses./urlhttp.afewguyscoding.com/2011/12
[8] Benenson, Z., Kroll-Peters, O., & Krupp, M. (2012, September). /survey-mobile-devicesecurity-threatsvulnerabilities-defenses.
Attitudes to IT security when using a smartphone. In Computer Science [29] Davi, L., Dmitrienko, A., Egele, M., Fischer, T., Holz, T., Hund, R., &
and Information Systems (FedCSIS), 2012 Federated Conference on (pp. Sadeghi, A. R. (2012, February). MoCFI: A Framework to Mitigate
1179-1183). IEEE. Control-Flow Attacks on Smartphones. In NDSS (Vol. 26, pp. 27-40).
[9] Ophoff, J., & Robinson, M. (2014, August). Exploring end-user [30] Kim, I. (2012). Keypad against brute force attacks on smartphones. IET
smartphone security awareness within a South African context. In Information Security, 6(2), 71-76.
Information Security for South Africa (ISSA), 2014 (pp. 1-7). IEEE. [31] Kumar, S., & Phrommathed, P. (2005). Research methodology (pp. 43-
[10] Mylonas, A., Kastania, A., & Gritzalis, D. (2013). Delegate the 50). Springer US.
smartphone user? Security awareness in smartphone platforms. [32] Couper, M. P. (2005). Technology trends in survey data collection.
Computers & Security, 34, 47-66. Social Science Computer Review, 23(4), 486-501.
[11] Alani, M. M. (2017). Android Users Privacy Awareness Survey. [33] Granello, D. H., & Wheaton, J. E. (2004). Online data collection:
International Journal of Interactive Mobile Technologies (iJIM), 11(3), Strategies for research. Journal of Counseling & Development, 82(4),
130-144. 387-393.
488 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Autism Spectrum Disorders (ASD) is quite difficult To address this issue, computer-aided learning for
to diagnose using traditional methods. Early prediction of individuals with mental imbalance was created [1], [2].
Autism Spectrum Disorders enhances the in general Furthermore it was concluded that it would be progressively
psychological well- being of the child. These days, the research on valuable if this would assist in identifying three noteworthy
Autism Spectrum Disorder is performed at a very high pace than territories like social and relational abilities, unbending nature
earlier days due to increased rate of ASD affected people. One of reasoning and relational abilities. Despite the fact that it is
possible way of diagnosing ASD is through behavioral changes of hard to get a correct number of Autism cases and it is
children at the early ages. Structural imaging ponders point to generally recognized that the predominance has been
disturbances in various mind regions, yet the exact neuro-
expanding in the course of recent years. There are signs that
anatomical nature of these interruptions stays misty. Portrayal of
cerebrum structural contrasts in children with ASD is basic for
some ASD range issue might be ascribed to a mix of certain
advancement of biomarkers that may in the long run be utilized hereditary susceptibilities; for example, introduction to
to enhance analysis and screen reaction to treatment. In this mercury at basic formative stages and diminished capacity to
examination we use machine figuring out how to decide a lot of discharge mercury [3]. There are a few investigations that
conditions that together end up being prescient of Autism have appeared even environmental toxicity may play a major
Spectrum Disorder. This will be of an extraordinary use to role in Autism. Analysts are yet dealing with these clutters to
doctors, making a difference in identifying Autism Spectrum comprehend the shrouded manifestations and subtleties of
Disorder at a lot prior organize. Autism.
Keywords—Autism Spectrum Disorder (ASD); ASD screening Autism spectrum disorders (ASD) is a disability in the
data; ABIDE; machine learning socio - development which was first identified by Kanner.
Kanner characterized ASD by the features of difficulties of
I. INTRODUCTION social behaviours, limited repetitive interests, difficulties in
interaction and behaviours [4]. In [5], it is stated that ASD is a
Autism Spectrum Disorder (ASD) is a neuro-
disability related to development which progresses from
developmental issue that can be described by distinct issues
childhood to adult and will exist as long as they live. Some of
that may arise during social collaboration, correspondence and
the common distinctive features of ASD are lack in language
conduct. There are numerous other mental disorders like
and communication skills, inadequacy in social interaction,
Autistic disorder and Asperger's disorder which have similar
show of inappropriate behaviours, etc.
classification of symptoms. Depending upon the severity of
symptoms, it prevails in numerous forms from extremely In DSM 5, the grouping of pervasive developmental
gentle ASD to exceptionally serious ASD. ASD must be disorders as specified by DSM-IV-TR is modified by placing
analysed immediately so as to follow the advancement of the autism spectrum along with social, communication and
youngster and give legitimate treatment. An underlying restricted or repetitive behaviours [6]. In DSM-IV-TR criteria,
screening is done in order to check the development of the pervasive developmental disorders are classified as five
child. ASD cannot be analysed by utilizing a solitary separate disorders such as, autistic disorder, childhood
screening. A second screening test must be performed when disintegrative disorder, Rett‘s disorder, Asperger‘s disorder
the child is 2-3 years of age. It very well may be precisely and pervasive developmental disorder. Children diagnosed
analysed after the second screening test process. In the on- with ASD has been increasing as the years are progressing [7].
going years, researchers are working on it to anticipate it
precisely within 18 months of the child. There are a lot of A. Problem Statement
formal screening tools accessible for doctors to expand the Machine learning has been utilized to anticipate instances
exactness for assessing the formative status of the children. of youngster misuse utilizing organized information and
Anyway, just a significant number of doctors utilize those literary data. This has nonetheless, not been done frequently
accessible tools. And furthermore, this needs a periodical and scarcely ever been finished for ASD and building up a
screening way to track the development status of the child. choice emotionally supportive network that helps doctors with
the recognition of ASD, has hardly been done, which is
demonstrated by the absence of writing on the utilization of it.
489 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
a) Identifying the risk caused by ASD: The seriousness Algorithm Applied Accuracy Precision Recall
of ASD was resolved utilizing Modified Checklist for Autism Naïve Bayes 0.512 0.479 0.512
in toddlers, otherwise called the M-CHAT. This is a screening Support Vector Machine 0.493 0.475 0.493
apparatus which comprises of inquiries that must be replied by J48 (Decision tree) 0.541 0.524 0.541
the guardians. The objective is to find the preeminent Random Forest 0.507 0.489 0.507
classifier for an autism dataset through feature relevance d) Using deep learning algorithms and Resting state
analysis. The classification algorithm is also used for functional magnetic resonance imaging (rs-fMRI): The
predicting the threat level of autism [10, 11]. Among various classification of brain imaging data is stricter when using deep
classification algorithm applied, for example, BVM, CVM and learning algorithms than using supervised methods. On
MLR created high exactness of 95.21% utilizing Runs utilizing significant neural framework a mean game plan
Filtering and it likewise precisely grouped the test dataset. exactness of 70% and a precision somewhere in the range of
This is valuable for age group of 16-30 years yet the forecast 66% and 71% in individual folds was acquired [15,16]. An
was not exact in all cases. expansion of 5% in classification accuracy was acquired while
b) Auxiliary Imaging using Voxel-based Morphometry: using deep learning classification method instead of Support
The examination demonstrated that utilizing a Multivariate Vector Machine [17]. In spite of the fact that the ABIDE
Pattern Analysis(MPA) and voxel-based morphometry (VBM) dataset contains sensitive varieties, the deep learning methods
to order structural magnetic resonance imaging data obtained envelop such assortments and yield better outcomes over
from 24 children and young people with mental imbalance IQ machine learning algorithms. The neural patterns obtained
coordinated neurotypical participants was pertinent just to from the classification show an anti-correlation of brain
small data sets [12]. Multivariate Pattern Analysis (MPA), function between posterior and anterior areas of the brain or
which is a pattern recognition technique that is solely based on cerebrum work among foremost and back regions of the brain.
machine-learning, can be used to group information by
B. Data Exploration
isolating between at any rate two classes. In MPA, the groups
are distinguished with about 90% of accuracy based on gray The Modified Autism Checklist in Toddlers (M-CHAT) is
matter in the medial prefrontal cortex, posterior cingulate a validated developmental screening tool for children aged 16
to 30 months. It is intended to identify children who may
cortex (PCC), and bilateral medial temporal lobes which are
benefit from a more thorough evaluation of development and
all regions within the default mode network (DMN) [13].
autism. This helps to find the best autism dataset classifier by
c) Machine Learning Techniques: There are many analysing feature relevance and classification algorithm.
machine learning algorithms that can be used for Among the various classification algorithms used, algorithms
classification. The three popular classification algorithms used such asBVM, CVMand MLR which produced an accuracy of
are Random Forest, Naive Bayes and Support Vector 95.21 % using Runs Filtering method [18]. This method
machines .Apart from these algorithms we can use the java accurately classified the test dataset.
implementation of the C4.5 algorithm known as the J48
The Autism Brain Imaging Data Exchange (ABIDE)
algorithm [14]. Using these assorted algorithms makes sure
initiative has totalled practical and auxiliary brain imaging
that the outcomes are highly reliable and it additionally helps information gathered from research facilities around the globe
us in finding whether the algorithm is usable for not for the to quicken our comprehension of the neural bases of autism.
classification task. When these algorithms were applied to the Every gathering was made through the accumulation of
dataset that was isolated into two classes-ASD or No ASD, the datasets freely gathered crosswise over in excess of 24 global
results are as in Table I. brain imaging research centers and are being made accessible
Be that as it may, utilizing the extra data we can discover to examiners all through the world, predictable with open
how seriously the individual is influenced with ASD. This is science standards. Since the data set contains more than 1112
finished by detaching into four groups. They are No ASD, records, before the data can be used for our machine learning
Mild ASD, Moderate ASD and Severe ASD [14]. On utilizing process, the data must be cleaned. The data will be explored
similar characteristics and the previously mentioned machine after cleaning the ABIDE data to determine if co-occurring
learning algorithms the outcome got are as in Table II. conditions are present in the data. If there are any clusters that
could help us predict ASD, it will also be investigated [19].
So as to build this we apply the 1-away strategy to the J48
Support Vector Machine algorithm will be used on the data
calculation as it is the best one. Thus, the exactness is
set, to determine if there are attributes that seem strongly
expanded from 54.1% to 90.2%.
490 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
correlated to ASD. In this algorithm, each data item is plotted classes, either the individual was Normal, or the individual
as a point in n-dimensional space with each feature being a was Mildly Autistic or Autistic. This was performed by using
coordinate value. ‗n‘ denotes the number of features selected the attributes resolved in the data set step mentioned above.
for identification [20]. We have used the MATLAB software for accuracy
calculation.
III. PROPOSED WORK
B. Enhancement of Image
A survey was done on various methods of predicting
Autism Spectrum Disorder (ASD) using different machine Image enhancement is a process that is widely utilized in
learning algorithms. Based on the survey a conclusion is numerous image processing applications, to amplify the
drawn that the algorithm Support Vector machine (SVM) can quality of images. In MATLAB software, the image is
be used for identifying patterns from autistic brain images enhanced by converting the image into gray scale. The
[21]. Here we have used the region called ―Corpus Callosum‖ difference is seen in the below given images (Fig. 2(a),
to identify the differences between the autistic and non-autistic Fig. 2(b)).
brain images. The Corpus Callosum is a fibre bundle which C. Image Binarization
connects the left and right hemisphere of the brain. Using
SVM algorithm, there were striking differences noted in that Image binarization is the process of converting a pixel
particular region where the thickness of Corpus Callosum was image into a binary image. Here, we use two main functions,
either too thick or thin compared to those brain images one to normalize the gray scale image by defining a threshold
without ASD [21], [22]. Furthermore it is found that the value and the other to convert the indexed image to black and
autistic brain had a decreased white matter volume and larger white intensity. The binary image after thresholding is shown
ventricles. Using these observations will ensure our result in Fig. 3.
outcomes are more reliable with higher accuracy rates. The D. Segmentation
proposed architecture is shown in Fig. 1.
Image segmentation is the process of splitting images into
multiple fragments. This division into fragments is mostly
based on the characteristics of the pixels of the image. We
have applied fuzzy c-means (FCM) clustering to produce one
or more clusters of the given binary image. After FCM
segmentation, the following clusters (Fig. 5) are generated out
of which one is chosen.
A. Data Sets
The ABIDE dataset contains 1112 records. This includes
data from 539 individuals with ASD and data from 573 typical
controls (ages 7-64 years). The Data Processing Assistant for (a)
Resting-State fMRI (DPARSF), the Configurable Pipeline for
the Analysis of Connectomes (CPAC), the Neuro imaging and
the Connectome Computation System (CCS) are used for
Functional pre-processing. This large pre-processed data is
used as an input for the classification algorithm.
IV. RESULTS
Below, the results from machine learning are described.
We divide our data set into three different classes, namely
Normal, Mildly Autistic and Autistic. A total of 100 iterations
are performed to classify the individuals falling under the
three classes.
A. Machine Learning Results using Three Classes
(b)
We applied the Support Vector Machine classification
algorithm to the data set when it was divided into three Fig. 2. (a) MRI Scan of An individual before Enhancement. (b) MRI Scan of
an Individual after Enhancement.
491 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
(a)
E. Accuracy Evaluation
A support vector machine (SVM) is a supervised learning
algorithm that creates a hyper plane in between data sets to
classify the belongingness of the data to its appropriate class.
The maximum accuracy of different types of SVM with 100
iterations to classify the images into the aforementioned three
classes has been evaluated. The confusion matrix generated is
given in Fig. 4.
The accuracy in percentage for different types of SVM is
mentioned in Table III and Fig. 6. (b)
(c)
(d)
Fig. 6. (a) Rapid Basis Function (RBF) Accuracy versus Regular Machine
Learning Techniques, (b) Linear Accuracy versus Regular Machine Learning
Techniques. (c) Polygonal Accuracy versus Regular Machine Learning
Techniques. (d) Quadratic Accuracy versus Regular Machine Learning
Fig. 5. Clusters Generated after FCM Segmentation. Techniques.
492 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. III. MAXIMUM ACCURACY ACQUIRED FROM VARIOUS TYPES OF [7] Centers for disease control and prevention (CDC), available at:
SVM https://www.cdc.gov/ncbddd/autism/data.htm
[8] Anibal Slon Heinsfeld , Alexandre Rosa Francop, R. Cameron
Types of SVM Normal Mildly Autistic Autistic Craddockt,, Augusto Buchweitzp,, Felipe Meneguzzio ―Identification of
RBF Accuracy 91% 84.5% 85.5% Autism Spectrum Disorder using Deep Learning and the ABIDE
Dataset―,2017.
Linear Accuracy 92.5% 91.5% 91.5%
[9] Felix D.C.C Beacher ,Eugenia Radulescu ,Ludovico Minati, Simon
Polygonal Accuracy 81.5% 87% 93.5% Baron-Cohen, Michael V.Lombardo, Meng-Chuan Lal, Anne Walker,
Dawn Howard, Marcus A.Gray ,Neil A. Harrison, Hugo D. Critchley
Quadratic Accuracy 84% 91% 93.5% ―Sex Differences and Autism: Brain Function during Verbal Fluency
and Mental Rotation‖,2012.
V. DISCUSSION AND CONCLUSION [10] Lucina Q. Uddin, Vino Menon, Christina B. Young, Srikanth Ryali,
Tianwen Chen, Amirah Khouzam, Nancy J. Minshew, and Antonio Y.
The objective was to identify the conditions that Hardan; ―Multivariate searchlight classification of structural MRI in
demonstrate to be prescient of ASD. This data can be utilized children and adolescents with autism‖,2017.
by physicians to enable them to confirm a complete formal [11] LamyaaSadouk ,TaoufiqGadi,andEl Hassan Essoufi; ―A Novel Deep
screening for ASD. Complex arrange parameters were utilized Learning Approach for Recognizing Stereotypical Motor Movements
to plan and analyze discriminate examination along with within and across Subjects on the Autism Spectrum Disorder‖,2018
bolster vector group of classifiers with a most extreme [12] M. S. Mythili and A. R. Mohamed Shanavas; ―An improved autism
reachable exactness of 94.7% utilizing four highlights and a predictive mechanism among children using fuzzy cognitive map and
feature extraction methods (feast)‖,2016
second request polynomial bit in SVM. The investigation has
[13] Christine Ecker, Susan Y Bookheimer, Declan G M Murphy;
endeavored to characterize the chemical imbalance range ―Neuroimaging in autism spectrum disorder: brain structure and function
scatter and creating subjects utilizing administered learning across the lifespan‖,2015.
systems as depicted in Figure 6. For future work, the focus is [14] Bram van den Bekerom ―Using Machine Learning for Detection of
towards the investigation of likelihood by utilizing profound Autism Spectrum Disorder‖,2017.
learning approaches for the programmed acknowledgment of [15] R.Geetha Ramani, K.Sivaselvi ―Autism Spectrum Disorder
SMM practices inside and crosswise over subjects. Identification Using Data Mining Techniques‖,2017.
[16] M. S. Mythili, A. R. Mohamed Shanavas; ―A Study on Autism Spectrum
REFERENCES
Disorders using Classification Techniques ―, 2014.
[1] Kathleen M. Carroll, Bruce J. Rounsaville, ―Computer-assisted Therapy
[17] Tabtah, F. ―Autism Spectrum Disorder Screening: Machine Learning
in Psychiatry: Be Brave—It‘s a New World‖, Current Psychiatry
Adaptation and DSM-5 Fulfillment.‖ Proceedings of the 1st
Reports, 2010 Oct; 12(5): 426–432.
International Conference on Medical and Health Informatics ,pp.1-6.
[2] Jorn Moock, ―Support from the Internet for Individuals with Mental Taichung City, Taiwan, ACM.,2017.
Disorders: Advantages and Disadvantages of e-Mental Health Service
[18] Thabtah, F. ―Machine Learning in Autistic Spectrum Disorder
Delivery‖, Frontiers in Public Health. 2014; 2: 65.
Behavioural Research: A Review.‖ To Appear in Informatics for Health
[3] Amy E. Kalkbrenner, Rebecca J. Schmidt, Annie C. Penlesky, and Social Care Journal. December, 2017 (in press),2017.
―Environmental Chemical Exposures and Autism Spectrum Disorders:
[19] Thabtah F, Kamalov F., Rajab K ―A new computational intelligence
A Review of the Epidemiological Evidence‖, Current Problems in
approach to detect autistic features for autism screening.‖ International
Pediatric and Adolescent Health Care. 2014 Nov; 44(10): 277–318.
Journal of Medical Infromatics, Volume 117,2018.
[4] Rachel Cooper, ―Diagnostic and Statistical Manual of Mental Disorders
[20] Christina Schweikert, Yanjun Li, David Dayya, David Yens, Martin
(DSM), Encyclopedia of Knowledge Organization, ISKO, available at:
Torrents, D. Frank Hsu ―Analysis of Autism Prevalence and
https://www.isko.org/cyclo/dsm.htm
Neurotoxins Using Combinatorial Fusion and Association Rule
[5] Vihang N. Vahia, ―Diagnostic and statistical manual of mental disorders Mining‖, 2009.
5: A quick glance‖, Indian Journal of Psychiatry. 2013 Jul-Sep; 55(3):
[21] Koyamada S, Shikauchi Y, Nakae K, Koyama M, Ishii S (2015) ―Deep
220–223.
learning of fMRI big data: a novel approach to subject-transfer
[6] Murat Gök, ―A novel machine learning model to predict autism decoding‖, 2015.
spectrum disorders risk gene‖, Neural Computing and Applications,
[22] Plis SM et al. (2014) Deep learning for neuroimaging: a validation study
2018, pp. 1-7.
frontiers in neuroscience 8,2014.
493 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Internet of Things (IoT) plays a pivotal part in our All these devices that enable them to upload input as well
mundane daily life by controlling electronic devices using as output to the Internet using cloud provisioning. The
networks. The controlling is done by minutely observing the information thus garnered is accessible for monitoring and
important parameters which generate vital pieces of information analysis anywhere in the globe via the internet [3]. In order to
concerning the functioning of these electronic devices. cut down on human effort and involvement, of late people
Simultaneously, this information will transmit these vital increasingly depend on embedded systems to control and
statistics from the transmitting device as well as save the same on monitor the factors affecting the ecosystem. Temperature and
the cloud to access by the applications and supplementary humidity are vital in observing and understanding nature. IoT
procedures to use them. This scrutiny associates the outcomes of
comes into the picture here by significantly enhancing the
the environmental observances like the humidity and
temperature measurements using sensors. The gathered
efficiency of the mechanism and systematically cutting down
information could be profitably used to produce actions like on human involvement, and thereby overall expenditure [4].
distantly dominant cooling, heating devices, or long term Practically, every part of exercise contains controlled
statistics, which will be useful to control the same. The detected schedules of temperature as well as humidity. However, the
data are uploaded to the cloud storage through network and exact value of temperature with its significant feature in any
associate using android application. The system employs Arduino field is essential in monitoring [5]. Constant perception in
UNO with Raspberry Pi, HTU 211D sensor device, and an
temperature is utilized in various industries like the
ESP8266 Wi-Fi module. The experimental results show the live
pharmaceutical industry as the driving force behind these
temperature and humidity of the surroundings and the soil
moisture of any plant using Arduino UNO with Raspberry Pi. monitoring systems, computerized and straightforward
Raspberry Pi is mainly used here for checking the temperature temperature sensors can use [6]. Resistors, semiconductors,
and humidity through the HTU 211D sensor element. The thermistors estimate temperatures values. These components
sensors are used for measuring the temperatures from the are present inside the sensor to retrieve the temperature in
surroundings, storing displayed information with different consonance with the circumstances. The primary goal of our
devices. Here, the ESP8266 Wi-Fi module has been used for data system is to supervise the live temperature and humidity within
storing purpose. a low cost [7].
Keywords—IoT; Raspberry Pi; Arduino UNO; data Raspberry Pi is the observational system or controller
transmission; sensors which is used for the cloud saving. Python is the programming
language which is utilized in Raspberry Pi. HTU 211D sensors
I. INTRODUCTION is a temperature sensor which is used here for the sensing
purpose [8]. This comprises of temperature ascertaining
IoT is used for connecting the electronic devices with the
capacity and favorable fundamental position of utilizing HTU
internet. The devices may vary from the temperature measuring
211D sensors, which boasts of less weight and ease of use. The
equipment and vehicles SOS system to other electronic devices
sensor is associated with Raspberry Pi utilizing connecting
such as sensors, software‟s, and network connectivity facilities,
wires. Temperature sensor HTU 211D sensors is utilizing is
which sanction collecting and exchanging data. The twenty-
perused put away, and shown in the Raspberry Pi unit [9].
first century has witnessed a massive paradigm shift to and
focusing on global attention onto IoT as a burgeoning IoT based devices in homes and industries are used for
discipline with multiple possibilities and diverse opportunities controlling all the electrical or electronic devices which are
for growth and development [1]. Internet connection facilitates present. Additionally, the saved information of the IoT devices
the smooth functioning of the devices that have become can be controlled from anywhere [10]. The sensor analyzes the
indispensable parts of our day-to-day lives and existence. The graphical representation of the observed data in every user-
Internet offers the provision to link and network different kinds defined format wherever in the world. In this work, IoT based
of devices like sensors and fitness devices. In the changed Arduino with Raspberry Pi microcontroller is used. Humidity
scenario post the September 11, 2001 attack on the United and temperature monitoring using Arduino is an exciting and
States where surveillance has gained paramount importance in secure process. Furthermore, this flexible system obtains more
proposed model security and survival, the internet facilitates values in calculating the actuator from the data saved on the
wholesome and perfect monitoring systems using closed- internet [11]. For connecting the Arduino board with Raspberry
circuit cameras [2]. Pi, USB line serial interface is essential to connect with any
application [12].
494 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
495 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 2 shows the solar panel power supply supplied to the mega 2560 Raspberry Pi microcontroller is the sole of this
whole system. During periods of rain, the battery saved data is arrangement, as well as the used power supply in this work, is
supplied. So, this will be beneficial for the farmers in the solar panel as a replacement for the regular home power
monitoring the agriculture area. Fig. 3 shows the circuit supply. Because the solar panel installation only required some
diagram for measuring the humidity and temperature of an amount; after the installation, there is no money spending
area. It shows the solar power supply of the system with an required. Next to the controlling process of such arrangement,
Arduino MCU with Raspberry Pi microcontroller, HTU 211D in addition to that, it can switch on and off the DC motor
sensor, and an ESP8266 Wi-Fi module. support on the soil water content level as well as the weather
season. Furthermore, the system is convenient for the weather
Temperature detector which is used here is a 4-pin low forecast.
price extremely reliable detector named HTU 211D
SENSORS. The first pin is connected with a Vcc node point. Then the collected data are transferred to the farmers live
Here, the utilized power supply is a solar panel. The second pin through the GSM to their cell phones. Based on the water level
is an information pin which will collect all the information measuring system, the collected data are sending to the
from outside and provides information to the microcontroller. farmer‟s cell phone continuously. They can switch on or off
The sensor pin configuration of HTU 211D sensor detectors is their motor based on the collected data from the water level
represented in Fig. 3. measuring system. This is beneficial for the farmers to control
the motors as well as can watch their plants from their house.
The temperature detector is very much useful for getting Moreover, this helps the plants from the overwatering. The
digital signal output. The HTU 211D SENSOR detector system is beneficial for water scarcity problems. The values are
includes a resistive wetness component and is connected to an uploaded within the stipulated time period through the ESP-
extraordinarily high-performing-8-bit-microcontroller. This 8266 Wi-Fi system. Then, from the Cloud, the humidity and
sensor provides the best worth output, fast response, low cost, temperature standards are measured using one ThingSpeak
and is interference-proof. Their temperatures vary from 00C- platform from anywhere.
550C, and the wetness value is among 20-90%.
Here, the used open data platform source is ThingSpeak
To transfer the readings of the device from HTU 211D software, which is free. Two-parameter tabs as humidity and
sensor to open the supply cloud ThingSpeak software, Arduino temperature are selected in Fig. 5. After the new channel log in
UNO with Raspberry Pi interfaces at the output with a LAN two API keys are generated. The original String API Key is
module named ESP-8266. In this module, ESP-8266 LAN “NTIM1RXET6YVUVWF”. Then replace the above line with
semiconductor device is connected with a full TCP/IP protocol the given program API key. Next, substitute the Host Name
stack. A voltage of 3.3V is ideal, which is then connected by and Password with Wi-Fi name and Wi-Fi password. The
Arduino UNO with Raspberry Pi on PC. The calculation is original String-Hostname is “Jonah” and the password is
performed using the AT command and wants the desired “2569696”.
sequence, to be used as a saver. The module will work on each
saver and server. It gets associated when connected to LAN
through the module so that it can transmit over the web.
During the testing of the ESP-8266 module, the module is
connected with the Arduino UNO Raspberry Pi. Then the
programmed Arduino UNO Raspberry Pi set up is connected
with the ThingSpeak platform through the ESP8266 LAN
module. ESP8266 LAN module acts as a protocol shopper, and Fig. 2. Solar Panel.
it will send the knowledge to ThingSpeak server. ThingSpeak
is the best IoT platform used for data collection and storage
purposes. Another unique feature of ThingSpeak is the data
analysis and comparison module. Comparison between two
different days can be accomplished using ThingSpeak
platform.
The HTU 211D sensors are used to senses the humidity and
temperature, and transfer the collected data through the 5th pin
of Arduino MCU connected Raspberry Pi, as shown in Fig. 4.
This set up can also control the DC fan, motor, and water
levels for supporting farmers. Then the measured values of
humidity and temperature values from the Arduino MCU are
uploaded to the Cloud.
Fig. 4 is the overall hardware setup of the proposed system.
This arrangement the Arduino mega 2560 Raspberry Pi
microcontroller is worn to design for calculating the
temperature, water level, and the humidity levels of the
primary agricultural areas also help the farmers. The Arduino Fig. 3. Adafruit HTU 211D Sensors.
496 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 6 shows the overall setup with all sensors and cloud. polls the device to retrieve information and sends over the web
The figure clearly shows the cloud, humidity, and temperature to ThingSpeak.
measurement systems. The program should be verified with the
Wi-Fi setup. To import the DHT library in Arduino Integrated To begin with, Raspberry Pi should be ready, and for that,
Development Environment (IDE), select the input sketch from need NOOBS. It is a software-based system manager that
the selected input folder. Then click „import‟ to retrieve the simplifies transfer, install, and then acquired wind of Raspberry
data from the library. To save cloud in the library, click „add Pi. Boot the NOOBS system once in the beginning; then get a
library‟; then select the library that has downloaded. Compile variety of operating systems (OS) to decide on from the
the sketch/program and upload to Arduino MCU through system. NOOBS makes obtaining started with Pi simple and
Arduino IDE. For these steps, better internet connectivity is includes a bunch of in OS to decide. The Raspberry Pi itself
indispensable, and hence, it should ensure beforehand. does not go together with the software system. Raspbian is the
“official” software system of the Raspberry Pi. Raspbian has
The central unit may be a microcontroller (Arduino UNO) been the quality Raspberry Pi in OS like UNIX.
and acts as the central processor unit for the complete system.
This unit interfaces with the device chip as the input for Since the system includes temperature and humidity
receiving temperature and humidness readings. For output, it watching, one device interface is required and no native storage
interfaces with the Wi-Fi module to send the received of information. Designate Arduino UNO with Raspberry Pi
information to the cloud over the web. The microcontroller microcontroller that serves the purpose well because of its
simplicity, lustiness, and low value. Fig. 6 shows an image of
497 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Arduino UNO with Raspberry Pi microcontroller utilized in the IV. RESULTS AND DISCUSSION
proposed system. This microcontroller board is predicated on IoT based temperature and humidity measurement system
the ATmega328P. The controller has a USB port, 14 digital provides an economical and safe system. This is very useful for
input/output pins, 6 analog input pins, 16 megacycle quartz the detection of agricultural-related parameters. The results of
with a power jack, and a button. It is battery-powered with a the temperature and humidness will see on the Raspbian OS
battery. It is programmable with the Arduino IDE via a sort B terminal. The central hardware element of the proposed system
USB cable. is the microcontroller that interfaces with alternative elements
The all humidity and temperature values will be uploaded of the system. Since the system includes temperature and
on the ThingSpeak platform. After that, one can see its humidity controlling that one device interface is required and
graphical representation of both humidity and temperature no primary storage of information. This designated an Arduino
values in a separate view window, as shown in Fig. 7. If one UNO with Raspberry Pi microcontroller.
wishes to change the channel or field name, one can change it In this regards, the Arduino mega 2560 Raspberry Pi
from the channel settings. Finally, the collected data are microcontroller‟s controlling the temperature, water level, and
transferred to the farmers time to time with the GSM to their the humidity levels measurement plots are plotted. The
cell phones. This is beneficial for the farmers to control the Arduino mega 2560 Raspberry Pi microcontroller used the
motors as well as can watch their plants from their house. The solar panel instead of the proposed electric supply. The
proposed hardware prototype is shown in Fig. 17. comparisons between the supply usage and their advantages
are studied. Moreover, the graphs of the controlling process
system, switch on and off the DC motor based on the soil water
content level and the weathers forecasting are plotted. Water
supply content levels, as well as the rainy season are drawn in
the ThingSpeak software.
Fig. 8 shows the water level checking setup. In this setup,
the sensor first senses the water level. Then data is then
transferred to the farmers live through the GSM to their cell
phones. This is beneficial for the farmers to control the motors
as well as can watch their plants from their houses.
Fig. 9 shows the comparison between the solar panel
powers to standard power. In this work, instead of the standard
power supply, a solar panel supply is used. The ThingSpeak
software attains the generated power difference to standard
power.
Fig.10 graph shows the humidity results of the proposed
system. Humidity is the quantity of water vapor present in the
Fig. 6. Overall Set up with All Sensors and Cloud. air. The water vapor amount is essential to attain the saturation
state in proportion with the increase in temperature value. As
the temperature of a parcel of air is low, it will eventually reach
the saturation point without adding or losing the mass of water.
The quantity of water vapor enclosed in the air can vary
significantly.
Fig. 11 shows the comparison graph between the
temperature and humidity levels every two hours. This is
essential for the system. These comparisons are used for the
weather forecasting unit of the proposed model. The
temperature and humidity levels are proportional to each other.
498 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
499 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 14. Graphical view of the Temperature Measurements Every Two Hours. Fig. 18. Constant Temperature Maintenance.
500 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Blockchain IoT
Decentralized Centralized
Resource consuming Resource restricted
501 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In the future, the extensive Arduino system can put into [11] Jiansheng PENG W.L, Qiwen HE, Design of smart home system based
practice as agriculture automation system and weather-based on the wireless MCU CC2510, Journal of Hechi University, 10, 2008.
fertilizer flower and monitor the value of the plants' growth via [12] Patricio G, Gomes L, Smart house monitoring and actuating system
development using automatic code generation, Industrial Informatics,
the mobile application. IoT based systems are a vital step in 7th IEEE International Conference, 256-261, 2009, 23-26. Vinay Sagar
sympathetic, relevance growth, accomplishment, and serve as a K.N, Kusuma S.M, Home Automation using Internet of Things, IRJET,
construction block for a numeral of practical modernization 02,2015.
technique controller. [13] Girish Birajdar “Implementation of Embedded Web Server Based on
ARM11 and Linux using Raspberry PI” International Journal of Recent
ACKNOWLEDGMENT Technology and Engineering (IJRTE) ISSN: 2277-3878, Volume-3
Issue-3, July 2014.
The author thanks King Abdulaziz University for this work. [14] Roselle B. Anire et al., “Environmental Wireless Sensor Network using
REFERENCES Raspberry Pi 3 for Greenhouse Monitoring System”, IEEE 9th
[1] Bhargav Goradiya, and H. N. Pandya, “Real time Monitoring & Data International Conference on Humanoid, Nanotechnology, Information
logging Systemusing ARM architecture of Raspberry pi & Ardiuno Technology, Communication and Control, Environment and
UNO” International Journal of VLSI and Embedded Systems-IJVES. Management (HNICEM), 2017.
ISSN: 2249 – 6556. Vol 04, PP: 513-517, July 2013. [15] KonstantinosTzortzakis, et al., “Wireless Self Powered Environmental
[2] M. Rahaman Laskar, R. Bhattacharjee, M. Sau Giri, and P. Monitoring System for Smart Cities based on LoRa”, Panhellic
Bhattacharya, “Weather Forecasting using Arduino Based Cube-Sat”, Conference on Electronics and Telecommunications (PACET), 2017.
Twelfth International Multi-Conference on Information Processing [16] Munsyi et al., “An Implementation of Data Exchange Using
(IMCIP) – 2016. Authenticated Attribute-Based Encryption for Environmental
[3] Vinayak Appasaheb Pujari, M. M. Raste, and A. A. Pujari, “Cost Monitoring”, 2017 International Electronics Symposium on Knowledge
Effective Automatic Weather Station-a Review”, International Journal Creation and Intelligent Computing (IES-KCIC).
of Electrical and Electronics Engineers (IJEEE)-Vol. No. 8 Issue 01, [17] Somansh Kumar, “Air Quality Monitoring System Based on IoT using
January-June 2016. Raspberry Pi”, International Conference on Computing, Communication
[4] C. H. Chavan, and V. Karande, “Wireless Monitoring of Soil Moisture, and Automation (ICCCA 2017).
Temperature and Humidity using Zigbee in Agriculture”, International [18] Cho ZinMyint, Lenin Gopal et al., “WSN-based Reconfigurable Water
Journal of Engineering Trends and Technology (IJETT)-Volume 11 Quality Monitoring System in IoT Environment”, 2017 14th
Number 10 – May 2014. International Conference on Electrical Engineering or Electronics,
[5] Mayur Randhir, R. R. Karhe, “Monitoring Of Environmental Parameters Computer, Telecommunication and Information Technology.
by Using Cloud Computing” International Journal of Computer Science [19] SanketSalvi, Pramod Jain et al., “Cloud Based Data Analysis and
Trends and Technology (IJCST) – Volume 3 Issue 3, PP: 151-155. May- Monitoring of Smart Multi-level Irrigation System Using IoT”
June 2015. International Conference on I-SMAC (IoT in Social, Mobile, Analytics
[6] Nelson Gonzalez, Charles Miers, Fernando Red´ıgolo, Marcos and Cloud), 2017.
Simpl´ıcio, Tereza Carvalho, Mats N¨aslund and Makan Pourzandi, “A [20] Alif Akbar Pranata, Jae Min Lee et al., “Towards an IoT-based Water
quantitative analysis of current security concerns and solutions for cloud Quality Monitoring System with Brokerless Pub/Sub Architecture”
computing” Journal of Cloud Computing: Advances, Systems and IEEE Transaction on Instrumentation and Measurement 2017.
Applications 2012, 1:11. [21] HakanUcgun, et al., “Arduino Based Weather Forecasting Station”, 2nd
[7] Mahesh D. S, Savitha S, and Dinesh K. Anvekar, “A Cloud Computing International Conference on Computer Science and Engineering 2017.
Architecture with Wireless Sensor Networks for Agricultural [22] J.Cabra, D.Castro et al., “An IoT approach for Wireless Sensor
Applications”International Journal of Computer Networks and Networks applied to e-health environmental monitoring”, 2017 IEEE
Communications Security Vol.2, No.1, January 2014, 34–38 Available International Conference on Internet of Things and IEEE Green
online at: www.ijcncs.org ISSN 2308-9830. Computing and Communications and IEEE Cyber, Physical and Social
[8] C. H. Chavan, and P. V.Karande, “Wireless Monitoring of Soil computing and IEEE Smart Data.
Moisture, Temperature & Humidity Using Zigbee in Agriculture” [23] Pablo Velasquez, et al., “A low-cost IoT based Environmental
International Journal of Engineering Trends and Technology (IJETT) – Monitoring System. A citizen approach to pollution awareness”, 2017
Volume 11 Number 10 - May 2014. CHILEAN Conference on Electrical, Electronics Engineering,
[9] Basil Ahammed, Design & Implementation of Smart House Control Information and Communication Technologies (CHILECON).
Using LabVIEW, International Journal of Soft Computing and [24] Nikolas Vidakis et al., “Environmental Monitoring through Embedded
Engineering (IJSCE), 1 (6), 2012. System and Sensors”, 52nd International conference on Power
[10] Dingrong Yuan, Shenglong Fang, Yaqiong Liu, The design of smart Engineering Universities, 2017.
home monitoring system based on WiFi electronic trash, Journal of [25] Shirazi et. al.. “Damage identification using wireless structural health
Software, 9 (2), 2014, 425-428. monitoring system through smart sensor application”, International
Journal of Advanced and Applied Sciences 4.2. 2017: 38-43.
502 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Chagas disease is caused by the parasite distributed at the provincial level. Also, it is necessary to
Trypanosoma Cruzi and transmitted by the Vinchuca. Bolivia is identify the risk that each sector has, in order to assess the
the country with the highest prevalence in the South American needs for control, care and planning according to local needs.
region; for example, in 2015, there was a prevalence of 33.4%.
This disease causes severe intestinal and cardiac problems in the The conception of the study of space as a new perspective
long term, 30% of the cases register cardiac symptoms, and 10% for epidemiological analysis in order to identify possible
have alterations in the esophagus or colon. This research aims to epidemiological outbreaks and the development of the disease
analyze the relationship between environmental factors and as a global process in a population space is called panoramic
Chagas outbreaks in an area of Bolivia to identify the epidemiology. In [5] for example, the reality of Argentina is
environmental conditions in which the disease is transmitted, described regarding the occurrence of dengue outbreaks, in
using epidemiological, meteorological data and also order to carry out epidemiological surveillance for the control
environmental indexes extracted from Landsat 8 satellite images. of the disease-causing vector. In work an exploration of the
Through a Principal Components Analysis (PCA) of the problem is carried out from the discipline of panoramic
environmental indexes extracted from the satellite images and epidemiology to analyze the distribution, ecology, and
the meteorological information, has been found that the behavior of the vector. Besides, an exploration of techniques in
environmental conditions that have a correlation with the classification and image processing is developed, in order to
occurrence of cases are: temperature, relative humidity,
generate a temporal space dispersion model of the vectors.
visibility, Normalized Difference Soil Index (NDSI) and Modified
Predictive maps were developed, of the focal density of Aedes
Normalized Difference Water Index (MNDWI).
aegypti, based on environmental information derived from
Keywords—Trypanosoma Cruzi; Vinchuca; Landsat 8; PCA; SPOT 5 HRG1 high-resolution spatial images and images of
Normalized Difference Soil Index (NDSI); Modified Normalized average spatial resolution of surface temperature derived from
Difference Water Index (MNDWI) Landsat 5 satellite information. A predictive model of biweekly
aedic indices was generated, which was based on macro-
I. INTRODUCTION environmental information from Landsat TM and ETM
In the world, the environmental characteristics and climatic sensors, and vector monitoring and control information. Then,
changes of the different ecosystems have a decisive influence space-time epidemiological patterns and population parameters
on some diseases that affect man, especially those associated of the vector were estimated; the effectiveness of vector control
with vectors such as Chagas. measures during the outbreak was also estimated.
Chagas disease is one of the most worrisome vector Also in [6] as part of the investigation of CONAE (National
diseases in Latin America; the World Health Organization Commission for Space Activities) and the Ministry of Health
(WHO) declares it as one of the most critical public health of Argentina, a project was developed to implement a Dengue
problems in America. Early Warning computer system, which calculates Dengue's
environmental risk in Argentina. The risk is assessed based on
According to [1] since 1990, significant success has been a static component related to historical environmental
achieved in the control of the parasite and the vector in Latin conditions and a dynamic component related to current
America. [2] In the last 20 years, different Initiatives achieved environmental conditions. The stratification of the localities of
a reduction in the transmission by domestic vectors as well as Argentina is shown as a result according to their environmental
blood transfusions. The burden of Chagas disease has been risk of Dengue.
reduced significantly (from around 30 million to 7 -8 million).
[3] Chagas disease is caused by the flagellated protozoan, The study published by Neteler [7] analyzes the
Trypanosoma Cruzi, which is transmitted to humans and other environmental conditions of the spread of the Asian tiger
mammals mainly through the droppings of blood-sucking mosquito Aedes Albopictus in Europe, as part of a constant
hemipteran insects on their host immediately after feeding. concern for public health due to outbreaks transmitted by
vectors. Data from the reconstructed daily satellite time series
The transmission of the disease to humans is due to the (MODIS Land Surface Temperature maps, LST) of
ability of the insect to explode and adapt to environments in northeastern Italy were used, as well as the annual average
different areas. The distribution of the disease is not uniform. temperature by areas, in order to reconstruct areas and compare
Previous studies like [4] show environmental differences in the them with the current known distribution of Ae. Albopictus in
situation of the problem of this disease. For the planning of northeastern Italy. LST maps show peculiar microclimatic
protection measures, it is essential to know how this disease is characteristics. From this data, surface maps are reconstructed
503 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
to predict the infection areas of the vector with an accuracy of Chagas can be transmitted through the consumption of
200m pixels. This is an important study that could be applied contaminated food, stinging of the infected vector, blood
to other arthropod species in which the temperature is a transfusion, transmission of the pregnant mother to the child,
relevant factor. transplantation of infected organs.
According to [8], the dispersal capacity of a disease that is The process of the disease has an incubation period of 4 to
transmitted by vectors depends on multiple environmental, 10 days, mostly asymptomatic. Then the three phases are
climatological, biophysical and social variables [1]. The distinguished, acute, indeterminate and chronic.
information extracted from the satellite images will allow us to
observe the climatic changes related to epidemiological The acute phase lasts between 2 to 4 months, is
outbreaks indirectly [8]. With satellite images of different asymptomatic, so the diagnosis is difficult at this stage, it can
spectral bands, different environmental parameters can be also present very mild symptoms. It is characterized by the
identified and measured [2]. high concentration of parasites in blood. Some of the
symptoms presented on occasion are: fever, headache, swollen
Images from the Landsat 8 satellite and the OLI-TRIS lymph nodes, slight pallor, muscle pain, breathing with
sensors will be used. With this analysis, we try to describe if difficulty, abdominal pain.
there are tools to generate models that describe the propagation
habitat of vectors that are the transmitters of endemic diseases. The indeterminate phase begins after the acute phase and
lasts for 8-10 weeks, regardless of the symptomatology. During
It is from these images that for certain period of time we this phase, the disease is usually asymptomatic. Although it
analyze the indices that are extracted from them to generate a still has quantities of the parasite in the blood, it is only
model that gives us a simplified representation of reality. This possible to diagnose it in 20% to 60% of the cases. It is during
model allows finding variables to understand the values of the this phase that the contagion becomes a great problem of
incidence and prevalence, as well as the process of public health, due to the ignorance of the presence of the
transmission of the disease in a certain period. disease.
The importance of this work is based on the fact that space During the chronic phase the parasite usually hides in the
is a privileged place. For this reason, satellite images can give cardiac and digestive organs, where they multiply and begin to
us an overview of the area that one wishes to analyze. The idea generate irreversible tissue damage. Only 30% of cases have
is to have global information about areas with a possibility of any cardiac symptoms, and about 10% have alterations in the
epidemiological outbreaks. All this information can contribute esophagus or colon. Moreover during this stage the amount of
to better decision making when planning prevention tasks and parasitaemia is low. The symptoms will depend on the damage
epidemiological management. of the parasite in the host organ. Chronic myocarditis is the
most common heart affection due to Chagas disease.
The limitations of the present study are that the temporal
resolution of the Landast 8 satellite is 16 days, which implies Among the symptoms that derive from the presence of the
having only one image to analyze per month, in addition in the parasite in the heart, are myocardial damage, arrhythmia and
study area there are not many weather stations to complement heart failure.
the data obtained from the satellite images.
Ventricular fibrillation is probably the most frequent
The content of this work is broadly divided into three main mechanism of sudden death in chronic Chagasic patients.
parts: in the first part the Chagas disease is described.
Afterwards a spatial temporal epidemiology analysis is The disease registers a percentage of mortality that
developed. Finally an analysis of the relationship between all oscillates between 55% and 65%, mainly due to cardiac
the data extracted is implemented. complications.
The treatment is especially useful in the acute phase of the
II. DESCRIPTION OF CHAGAS DISEASE disease, where it can cure up to 100% of cases. The
This section seeks to expose the basic aspects of Chagas effectiveness of the treatment decreases in an inverse manner
disease, besides shows the growing presence of Chagas in over time, because the parasite can cause irreversible damage
South America and mainly in Bolivia in recent years. to some organs.
A. Definition and basic Aspects of the Disease This treatment is based mainly on benznidazole and
The disease of American trypanosomiasis or Chagas, it is a nifurtimox capable of killing the parasite depending on the
type of zoonosis, a disease typical of the encounter with phase of infection.
animals of contact with people, typical of this continent, In some chronic cases, the treatment although it may not
because has existed in Latin America since before the kill the parasite completely, can prevent or delay the
conquest. progression of the disease, although it is important to consider
In 2018 WHO reported 6 or 7 million people infected in 21 the adverse consequences of dosing them over time.
countries by the parasite Trypanosoma Cruzi [3], which causes Unfortunately there is no vaccine against Chagas disease,
this disease and 90 million people are at risk of contracting the but the prevention method is the control of the vector in the
infection. areas of greatest incidence.
504 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The T. Cruzi can infect several species, depending on the Bolivia there are 21 types of vinchucas identified, the most
geography of the area, for this reason it is recommended for its common vinchuca is the Triatoma Infestans, which is
prevention the fumigation of areas of greater risk, cleaning and responsible for the largest number of Chagas (CH) cases
improvement in the conditioning of homes, hygiene, adequate recorded in the country.
conservation of the food and develop regular serological tests
in risk areas. The country with the largest dispersion area of the CH
vector (Triatoma Infestans) is Bolivia. [14] In the endemic
B. Presence of Chagas in Latin America vector map of the Ministry of Health Chagas disease is
Chagas is a parasitic disease that has become a public dispersed in approximately 60% of the Bolivian territory.
health problem in Latin America [9], because it has a presence In the country, three endemic zones were identified: the
in at least 21 countries in the region. valley area comprised by the departments of Cochabamba,
In America the infection is located from the south of the Chuquisaca, Tarija and Potosí, the Chaco area comprised by
United States to Argentina and Chile. According to [10] due to the departments of Santa Cruz, Chuquisaca and Tarija, and
this disease in this continent about 50,000 people die each year finally the Amazon area departments of Beni, Pando, part of
and more than 100 million people are at risk of becoming Santa Cruz, north of La Paz and north of Cochabamba.
infected. According Médecins Sans Frontières (MSF) in [15] the
In 2013 according to the World Health Organization [11], it South American region, Bolivia register more than 600,000
has been estimated that of the 8 million people carrying T. people infected with the disease. Besides, an average of 8,000
Cruzi, the largest number of cases are concentrated in Latin new cases of Chagas is registered each year.
America, although important number of cases are also The Chagas Prevention Program of the Ministry of Health
calculated in the United States, Canada, Spain, Japan and of Bolivia indicates in [13] that the country register the highest
Australia. prevalence in the South American region, for example in 2015
Chagas disease in Latin America can be associated to there was a prevalence of 33.4%.
multiple factors, including dwelling houses built with materials In 2016 the Ministry of Health of Bolivia recorded 17,892
such as adobe, mud and straw, this type of structures can be new cases, in the department of Santa Cruz, 57.72% of the total
seen in rural and suburban areas of the region. cases of Chagas were concentrated, followed only by the
There are around of 140 species of vinchucas in the world, department of Cochabamba, which concentrated a total of
of which the majority are distributed throughout the American 23.13% of infected persons [14].
continent, only a few species are present in Asia, Africa and III. SPATIAL TEMPORAL EPIDEMIOLOGY ANALYSIS
Australia. [11].
For this research it is necessary to select a study area of
In Latin America, together with the Pan American Health interest in Bolivia, acquire the different types of data and
Organization (PAHO) and the World Health Organization performs the treatment of the satellite images, this section
(WHO), a horizontal technical cooperation strategy was exposes these main procedures.
developed among countries to prevent and control Chagas
disease in the region. [3] For example, these initiatives have A. Selection of the Study Area
been developed in the Southern Cone (1992), the Andean Chagas has become an emerging disease in Bolivia, since it
countries (1998) and the Amazonian countries (2003), among is dispersed in approximately 60% of the Bolivian territory.
other cooperation strategies in the region. The CH vector (Triatoma Infestans) it has found in the warm
This cooperation has contributed to the elimination of departments of the country, such as Santa Cruz, Tarija,
allochthonous species of vectors, the detection of congenital Chuquisaca and Cochabamba, the adequate conditions to
cases, the reduction of prevalence in children and the prevail and transmit the disease.
improvement of the quality of treatment of infected and sick We can also observe that Chagas disease has presence in
people, among others. the nine departments of Bolivia. For this reason we must select
In the specific case of South America, the vector Triatoma among the most affected departments of the country by this
Infestans predominates in Argentina, Bolivia, Brazil, Chile, viral disease and where there is a high number of cases.
Paraguay, Peru and Uruguay. [12] Besides in Colombia and In order to select a suitable research study area the National
Venezuela the predominant vector is R. Prolixus. Program of Preventive and Control of Chagas in Bolivia was
According to figures calculated by WHO in 2010, Bolivia revised, where the epidemiological situation of the country in
is the country with the highest incidence of Chagas in South 2017 can be observed:
America. As we can observe in Table I, there is a high number of
C. Presence of Chagas in Bolivia Chagas cases in three departments: Santa Cruz, Cochabamba
and Tarija. Also we can observe a moderate number of cases in
According to the Institute of Development Health Research three departments: Potosi, Chuquisaca and Beni. Finally there
(IINSAD) of the Universidad Mayor de San Andrés (UMSA) is a low number of cases in three departments: La Paz, Pando
in [13] there are 140 species of vinchucas in the world, in and Oruro.
505 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. I. TOTAL CASES OF CHAGAS DISEASE IN BOLIVIA IN 2016 TABLE. II. TOTAL CASES OF CHAGAS DISEASE IN BOLIVIA IN 2017
In the nine departments of Bolivia there is a total of 15911 TABLE. III. INCIDENCE RATE AND POINT PREVALENCE OF CHAGAS
cases of Chagas that have been recorded throughout the year of
2017, as can be seen in Fig. 1. Incidence
Point
rate
Departments TOTAL Population prevalence
According to the collected data, the department of Santa (100,000
(%)
Cruz, is the department that has registered the most high inhabitants)
number of cases in 2017, with a total of 7777 cases, in second BENI 300 462,081 64.924 0.065
place is the department of Cochabamba with a total of 4794 CHUQUISACA 672 621,148 108.187 0.108
cases and in third place is the department of Tarija with a total COCHABAMBA 4,794 1,943,429 246.677 0.247
of 1457 cases of Chagas. LA PAZ 159 2,862,504 5.555 0.006
It can be observed that in eight of the Bolivian departments ORURO 11 531,890 2.068 0.002
the number of female cases is greater. The departments of PANDO 19 139,018 13.667 0.014
Santa Cruz and Cochabamba are those that report the highest
POTOSI 722 880,651 81.985 0.082
number of female cases.
SANTA CRUZ 7,777 3,151,676 246.758 0.247
In order to correctly identify the study area, the incidence TARIJA 1,457 553,373 263.294 0.263
rate and point prevalence of Chagas disease in the nine
departments of Bolivia were calculated in Table II.
As can be seen in Table III, the departments with the
highest incidence rate and prevalence percentage are: Tarija,
Santa Cruz and Cochabamba respectively. However, the
department of Santa Cruz has the highest number of cases,
because it has a larger population (more than three millions of
inhabitants), for this reason Santa Cruz is selected as the area
of analysis.
Since the department of Santa Cruz is the largest
department in Bolivia, is necessary to select only one specific
area to carry out the research. Fig. 2. Provinces of the Department of Santa Cruz.
The department of Santa Cruz has 15 provinces, as can be As can be seen in Table IV, the provinces with the highest
seen in Fig. 2, of which according to the Ministry of Health in number of cases in the department of Santa Cruz are: Andres
Bolivia 14 provinces have the presence of Chagas in 2017, the Ibañez (with 6,315 cases), Obispo Santistevan (with 811 cases)
information collected can be observed in Table III. and Warnes (with 330 cases).
As can be observed in Fig. 3, in 9 of the 14 provinces
affected by Chagas in Santa Cruz it can be seen a higher
number of cases in the female population, except in the
province of German Busch where it can be seen that the
number of cases is the same in the female and male population.
Andrés Ibañez province represents 81.9% of the total
number of cases registered in the department of Santa Cruz
throughout 2017, followed in the second place by the province
of Obispo Santistevan, which represents 11.2% of the cases
Fig. 1. Chagas Disease in the Bolivian Departments in 2017.
and in the third place is the province Warnes which represents
2.2% of the total cases, as can be seen in Fig. 4.
506 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. IV. CASES OF CHAGAS IN THE PROVINCES OF SANTA CRUZ IN 2017 Finally the province Warnes has two municipalities, which are:
TOTAL
Warnes and Okinawa. It is located at the coordinates:
17°20'00"S 63°00'00"W.
Province MALE FEMALE TOTAL
ANDRES IBAÑEZ 2,498 3,817 6,315 B. Data Acquisition
CHIQUITOS 11 46 57 For the present work, one needs three types of data on
CORDILLERA 14 18 32 which the analysis, correlations, and conclusions will be
obtained. These data are: satellite images, epidemiological
FLORIDA 1 4 5
data, and meteorological data. The description of each of these
GERMAN BUSCH 4 4 8 data is detailed below.
GUARAYOS 13 16 29
a) Satellite images: There are different satellites that
ICHILO 14 15 29
provide us multispectral images, from which it is possible to
MANUEL MARIA CABALLERO 5 2 7 collect environmental information, which in this case will be
ÑUFLO DE CHAVEZ 43 40 83 used to identify the environmental parameters that influence in
OBISPO SANTISTEVAN 341 470 811 Chagas outbreaks in three provinces of Santa Cruz in Bolivia.
SARA 31 19 50 For this research we collected information from the Landsat
VALLEGRANDE 5 12 17 8 satellite, launched on February 11, 2013, these satellite
images are downloaded for free from the web:
VELASCO 3 1 4
https://earthexplorer.usgs.gov/.
WARNES 67 263 330
The Landsat 8 satellite has two sensors, OLI and TIRS.
Among the most important features observed in [16], can be
mentioned that it is heliosynchronous, at an orbital height of
705 km, WRS-2 (Worldwide Reference System), with an
inclination of 98.2°, a temporal resolution of 16 days and
having a radiometric resolution of 12 bits, spatial resolution of
30m, and spectral resolution of nine bands 5 of them in the
visible field and the others in the non-visible field.
507 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The images selected for this research were compiled from The database compiled also contains data on the types of
the USGS website. It was identified that the scene that covers health facilities where cases of Chagas were reported in 2017.
83% of the study area, is in the location Path 231 and Row 72, It is important to mention that in Bolivia there are first, second
according to the WRS system. and third level hospitals, health centers and medical posts
among other smaller health facilities.
The scenes of all the months of the year 2017 were
analyzed, choosing only those that have a minimum cloud TABLE. V. SATELLITE IMAGE DATASETS OF THE STUDY AREA
percentage, the details can be observed in the following
Table V. Date % Cloud
Data set Path-Row
acquired coverage
Those images registering a minimum cloudiness of less LC08_L1TP_231072_20170119
than 40% were selected. The selected scenes have an average 231 - 72 19/01/2017 44.76
_20170311_01_T1
cloudiness of 21.80%, the selected months are: February, July, LC08_L1TP_231072_20170220
August, October and November. 231 - 72 20/02/2017 33.55
_20170301_01_T1
Of the five datasets selected, we have been downloaded: 11 LC08_L1TP_231072_20170308
231 - 72 08/03/2017 64.03
raster bands (GeoTIFF), the BQA file (16 bits quality control _20170317_01_T1
file), and the metadata file (MTL), in total 14 files. LC08_L1TP_231072_20170425
231 - 72 25/04/2017 69.88
_20170502_01_T1
The spatial subset of the 11 bands of the image is:
7622x7732 pixels, with a vertical and horizontal resolution of LC08_L1TP_231072_20170527
231 - 72 27/05/2017 71.7
_20170615_01_T1
96pp, and a depth of 32 bits. The image corresponding to band
8, is the only one that has a resolution of 15242x15242 pixels, LC08_L1TP_231072_20170612
231 - 72 12/06/2017 82.26
with a vertical, horizontal and depth values similar to those _20170628_01_T1
already mentioned. LC08_L1TP_231072_20170714
231 - 72 14/07/2017 10.9
_20170726_01_T1
b) Epidemiological data: The Ministry of Health of
LC08_L1TP_231072_20170831
Bolivia, through its Departmental Health Service (SEDES), by 231 - 72 31/08/2017 5.61
_20170915_01_T1
its acronym in Spanish “Servicio Departamental de Salud”, in
LC08_L1GT_231072_20170916
a joint effort of the nine departments of Bolivia, has developed 231 - 72 16/09/2017 100
_20170929_01_T2
and implemented the National Health Information System.
LC08_L1TP_231072_20171018
This information system contains fourteen-year data from 231 - 72 18/10/2017 20.27
_20171025_01_T1
2005 to 2018, also it contains different groups of variables, LC08_L1TP_231072_20171103
such as immuno-preventable diseases, sexually transmitted 231 - 72 03/11/2017 38.69
_20171109_01_T1
diseases and vector-borne diseases, among many others.
LC08_L1GT_231072_20171205
231 - 72 05/12/2017 100
The national information system has been developed to _20171222_01_T2
maintain an epidemiological surveillance program in all the
health establishments of the nine departments of Bolivia. The TABLE. VI. DESCRIPTION OF THE COLLECTED EPIDEMIOLOGICAL DATA
tool is available on the web portal of the Ministry of Health
Bolivia in the informatics tools section available at: Database Form Disease Year Months Weeks
https://snis.minsalud.gob.bo/. Notification for
Acute January to
epidemiological 302 a 2017 52
This informatic tool allows to perform an epidemiological surveillance
Chagas December
surveillance of vector-borne diseases, including: Dengue, Zika,
Chikungunya and Chagas disease among others. Monthly
notification for Acute January to
302 b 2017 52
The tool allows a search of data by departments, provinces, epidemiological Chagas December
municipalities, networks and health establishments throughout surveillance
Bolivia and in the different months of the year of interest.
For this study, the epidemiological database corresponding
to the 2017 period is used, as can be seen in Table VI.
In the database collected, as can be seen in Table IV, the
number of cases of acute Chagas reported in the study area is
obtained, but in addition these cases are classified by gender
and by age, from less than 6 months to over sixty years of age.
It can be observed in Fig. 6 that the age group with the
highest number of cases of Chagas is the group of 60 years and
over, with 2,560 cases. Moreover, the age group with the least
presence of Chagas is the group of six months to less than one
year, with 40 cases. Fig. 6. Age Groups of Chagas Cases in the Study Area.
508 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
As it can be observed in Fig. 7 the Chagas disease has been the shape of the area of interest, corrections must be made as
found in 11 types of health facilities, both public and private, part of the preprocessing stage and finally an algorithm must
belonging to the health network of the Bolivian state. It can be be programmed in MATLAB to extract the necessary
seen that there is a greater presence of Chagas in the information from the satellite images.
Ambulatory Health Centers, on the contrary it can be seen that
there is a lower presence of the disease in the poly-medical a) Preprocessing: To begin with the preprocessing of
centers. the images, the maps of departmental, provincial and
municipal limits were used in shape file format, selecting
Table VII is a summary of the epidemiological information those that correspond to the area of interest identified in the
extracted and shows the total cases of Chagas in three epidemiological analysis.
provinces selected as the study area. We can only observe the The shapefiles were downloaded from the website of
cases of Chagas of certain months, this is due to the cloud GeoBolivia, which were based on the maps created by the
coverage, as was previously explained we only select the Bolivian Geographical Institute, which has the information
months of the year with a low percentage of clouds. provided by the Ministry of Autonomies with data updated to
c) Meteorological data: The meteorological data are April 2015, this divides Bolivia into 339 municipalities.
essential since they are the data that we seek to equate with the These geographic information files were created on
environmental parameters that will be obtained from the April 15, and its publication as an available web resource
satellite images to correlate them with the epidemiological occurred on May 10, 2015, under the OGC protocol: WMS-
data. 1.1.1-http-get-map.
The meteorological data come from the meteorological
stations of the National Service of Meteorology and Hydrology After obtaining the shapefile of the area of interest, which
(SENAMHI) of Bolivia (http://senamhi.gob.bo/index.php as indicated above, comprises three provinces: Andrés Ibañez,
/inicio). Obispo Santistevan and Warnes. The next process was to crop
the 11 bands and the BQA raster of each multispectral image
The extracted data are daily precipitation, daily relative using the obtained shapefile, in each subset the spatial
humidity, and the daily visibility. resolution of the input satellite image is preserved.
In Table VIII, one can see a table with the collected
meteorological data for February, July, August, October, and
November of 2017.
The data collected corresponds to the year 2017 (which
corresponds to the year that is being analyzed in the present
work with the satellite images) in 852450 SLET
meteorological station, since this station is in the area under
analysis. One chose February, July, August, October, and
November because these months correspond to the satellite
images.
C. Treatment of Satellite Images
It is necessary to extract the environmental parameters of Fig. 7. Type of Health Establishments in the Study Area.
the satellite images, for this the images must be cropped with
TABLE. VII. CHAGAS CASES IN THE STUDY PROVINCES
509 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
510 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. XI. ENVIRONMENTAL INDEXES Obtaining results: In this stage the raster resulting from the
Name Definition Equation
calculation of each of the parameters must be obtained, an
evaluation of the results obtained must be carried out to
NDSI
Normalized Difference Soil corroborate that the obtained indexes are within the maximum
Index ranges established in the theory, to confirm that the
Normalized Difference calculations made are correct.
NDMI
Moisture Index
IV. ANALYSIS OF THE RELATIONSHIP OF ALL THE DATA
Normalized Difference EXTRACTED
NDVI
Vegetation Index
Among the obtained data, one has seven parameters or
Normalized Difference Water indices extracted from the satellite images, three parameters
NDWI
Index (Content in leaves) from the meteorological stations, and the data related to the
Modified Normalized cases of Chagas in the area under analysis. The parameters
MNDWI
Difference Water Index obtained from the satellite images and the data from the
meteorological station, both dataset form a total of ten features.
The first analysis to be performed is the calculation of the
correlation between the ten selected features.
In Table XII, one can see the Pearson correlation between
the eight selected features where a good correlation is shown
between the NDMI and the NDWI. The p-value of the
correlations previously mentioned (see Table XIII) is 0.0001 in
both cases; this value being less than 0.05 confirms the
significance of both correlations.
Having ten features, one applies the principal component
analysis (PCA) to reduce the space of variables to be analyzed.
After the PCA, it is observed that the workspace is reduced
only to four principal components (see Table XIV).
Among the four main components shown in Table XV,
they explain the 100% of the data collected. As shown in
Table XII and Fig. 9, Component 1 explains more than 94% of
the data.
Since Component 1 is the one that practically explains the
data thoroughly, we look for the correlation of Component 1
with the epidemiological data that represent the number of
cases of Chagas in the area under analysis.
In Table XVI, it can be seen that there is a high correlation
(Pearson correlation) between Component 1 and
epidemiological data. This absolute value of correlation is
equal to 0.8469.
In Fig. 10, we can observe a correlation map between
Chagas cases and the results found in the analysis of the
principal components extracted from Table XIV.
The characteristics of component 1 obtained by PCA were
combined to create heat maps where the blue areas represent
high values and the orange areas correspond to low values.
Fig. 8. Flow Diagram of the Processing Algorithm.
511 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. XII. CORRELATION BETWEEN ALL TEN FEATURES FROM SATELLITE IMAGES AND METEOROLOGICAL STATION
NDSI NDMI NDVI NDWI (Leaves) MNDWI Temperature Relative humidity Visibility
NDSI 1,0000 -0,9976 -0,5035 -0,9976 -0,6463 -0,1595 0,4019 -0,0570
NDMI -0,9976 1,0000 0,4551 1,0000 0,6818 0,1573 -0,4033 0,1024
NDVI -0,5035 0,4551 1,0000 0,4551 -0,3268 -0,3267 -0,0563 -0,0513
NDWI (Leaves) -0,9976 1,0000 0,4551 1,0000 0,6818 0,1573 -0,4033 0,1024
MNDWI -0,6463 0,6818 -0,3268 0,6818 1,0000 0,4816 -0,3471 0,0405
Temperature -0,1595 0,1573 -0,3267 0,1573 0,4816 1,0000 -0,7676 -0,7429
Relative humidity 0,4019 -0,4033 -0,0563 -0,4033 -0,3471 -0,7676 1,0000 0,4133
Visibility -0,0570 0,1024 -0,0513 0,1024 0,0405 -0,7429 0,4133 1,0000
NDSI NDMI NDVI NDWI (Leaves) MNDWI Temperature Relative humidity Visibility
NDSI 1,0000 0,0001 0,3872 0,0001 0,2387 0,7978 0,5024 0,9274
NDMI 0,0001 1,0000 0,4412 0,0000 0,2049 0,8006 0,5008 0,8698
NDVI 0,3872 0,4412 1,0000 0,4412 0,5915 0,5915 0,9284 0,9347
NDWI (Leaves) 0,0001 0,0000 0,4412 1,0000 0,2049 0,8006 0,5008 0,8698
MNDWI 0,2387 0,2049 0,5915 0,2049 1,0000 0,4115 0,5671 0,9484
NDBI 0,0001 0,0000 0,4412 0,0000 0,2049 0,8006 0,5008 0,8698
UI 0,0003 0,0018 0,3208 0,0018 0,2921 0,7973 0,5077 0,9971
Temperature 0,7978 0,8006 0,5915 0,8006 0,4115 1,0000 0,1297 0,1503
Relative humidity 0,5024 0,5008 0,9284 0,5008 0,5671 0,1297 1,0000 0,4892
Visibility 0,9274 0,8698 0,9347 0,8698 0,9484 0,1503 0,4892 1,0000
TABLE. XIV. RESULT OF THE PCA. THE WORKSPACE IS REDUCED TO FOUR PRINCIPAL COMPONENTS
TABLE. XV. PERCENTAGE OF EXPLANATION OF THE FOUR PRINCIPAL COMPONENT AFTER THE PCA
Explained (%)
Component 1 94.125
Component 2 5.475
Component 3 0.396
Component 4 0.004
TABLE. XVI. CORRELATION BETWEEN THE COMPONENT 1 AND THE EPIDEMIOLOGICAL DATA
Pearson
Correlation -0.8469
512 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
513 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[12] F. M. Sano, M. Sano, y O. P. de la Salud, «La enfermedad de Chagas. A Landsat 8 con el Módulo Flaash del Software ENVI. Rev Geográfica
la puerta de los 100 años del conocimiento de una endemia ancestral», A América Cent 2014; 2: 39–59.
doença de Chagas. Perto dos 100 anos do conhecimento de uma endemia [19] Tardy B, Rivalland V, Huc M, et al. A software tool for atmospheric
americana ancestral, 2007. correction and surface temperature estimation of landsat infrared thermal
[13] C. Forsyth, «From Lemongrass to Ivermectin: Ethnomedical data. Remote Sens; 8. Epub ahead of print 2016. DOI:
Management of Chagas Disease in Tropical Bolivia», Med. Anthropol., 10.3390/rs8090696.
vol. 37, n.o 3, pp. 236-252, 2018. [20] Kruse F. Comparison of (ATREM), (ACORN), and (FLAASH)
[14] Ministry of Health Bolivia, “Boletín de vigilancia epidemiológica, Atmospheric Corrections using Low-Altitude (AVIRIS) Data of
Chagas Agudo”, 2016. Boulder, Colorado. 2004.
[15] A. Rassi, A. Rassi, y J. Marcondes de Rezende, «American [21] Remer LA, Tanré D, Kaufman YJ. Algorithm for remote sensing of
trypanosomiasis (Chagas disease)», Infect. Dis. Clin. North Am., vol. tropospheric aerosol from MODIS: Collection 005.
26, n.o 2, pp. 275-291, jun. 2012. [22] Deng Y, Wu C, Li M, et al. RNDSI: A ratio normalized difference soil
[16] Roy DP, Kovalskyy V, Zhang HK, et al. Characterization of Landsat-7 index for remote sensing of urban/suburban environments. Int J Appl
to Landsat-8 reflective wavelength and normalized difference vegetation Earth Obs Geoinformation 2015; 39: 40–48.
index continuity. Remote Sens Environ 2016; 185: 57–70. [23] Xu H. Modification of normalised difference water index (NDWI) to
[17] Chander G, Markham BL, Helder DL. Summary of current radiometric enhance open water features in remotely sensed imagery. Int J Remote
calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI Sens 2006; 27: 3025–3033.
sensors. Remote Sens Environ 2009; 113: 893–903. [24] Zha Y, Gao J, Ni S. Use of normalized difference built-up index in
[18] Aguilar-Arias H, Mora-Zamora R, Vargas-Bolaños C. Metodología para automatically mapping urban areas from TM imagery. Int J Remote
la Corrección Atmosférica de Imágenes Aster, Rapideye, Spot 2 y Sens 2003; 24: 583–594.
514 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In the fingerprint image processing, various identity theft. The unlimited capacity of fingerprint sensors as
methods have been suggested as using band pass filter, Fouries advantage related to conversion each fingerprint to an image,
transform filter and Fuzzy systems. In this paper, we present a where latterly it processed in the database. The portability of
useful and an applicable fingerprint security system for student’s the fingerprint sensors considered as advantage because of the
examination using image processing on such away and a well- small device with the tiny components.
organized algorithm is applied. As a university team work, we
have recently tested this security procedure for different samples The fingerprint-based system offers usually a reliability of
of students in our institution. The experimental results show a such systems with different challenges. Using an external
high level of accuracy is obtained. Due to the need to connect and database, however, affects the efficiency of fingerprint-based
manage the connection, we use the Ethernet card and the system. Furthermore, the commercial devices developed by the
Arduino Uno card which they are combined together in such a fingerprint have been constructed to store the fingerprint
way to do so. Moreover, the administrator runs a special website features for limited users. Those devices have been developed
in the PC to assign ID to the scanned fingerprint. The calculation also as an individual unit with their own interface and database,
of the proposed system is carried out by uploading a suitable once the user places the fingerprint confirmation message
Adafruit fingerprint library to the used Audruino Uno card. arisen on the output screen without other details (text or
Finally, the most important security point is that the PC has been password). Another challenge related installation process
used not only to send the developing software into the Uno card
where the commercial device has to be fixed in a place and
but also to disconnect the process electronically while the code is
connected to the facility networks, where the data transfer to
running.
and from the database server. This process increases the cost
Keywords—Finger-print; examination system; image and chance of hacking. The security web services developed
processing; bio informatics for the examination system represents another challenge. These
services could be a promise way to improve the security of
I. INTRODUCTION examination system once it addressed with biometric
The authentication plays an important role in examination processing such as fingerprint.
systems. The most authentication approach applied in those In this work, a novel secure fingerprint-based
systems is identity- based authentication. However, the reality authentication system has been developed to discuss these
in a lot of developed examination system is that the identity- challenges in context of web services. The developed system
based authentication is not enough to verify the student has been developed to further ensuring for usability,
identification. confidentiality and portability of such fingerprint based system.
One of authentication solutions is the biometric processing II. RELATED WORKS
based on iris or fingerprint data. Because of cost related issues,
fingerprint has been adopted as an automatic data analysis for Identity authentication considered a topic of interest in the
identification. The main difference between the password- recent years, while increasing the need for more reliable and
based authentication and fingerprint based authentication is useful identity authentication systems for security [1]. The old
that the first one cannot sometimes protect from unauthorized traditional authentication systems depend on using passwords
accessing to users data especially in examination based website or ID cards found to be less reliable [2]. Indeed, the traditional
applications. Using the fingerprint data does not mean that the authentication systems are not able to recognize the original
examination system avoids using the password-based person from the cheater using the password [1].
authentication. In addition to the password-based The case of how to improve the security in conjunction
authentication, the examination system has to be boosted with increase usability and decrease interventions still under
therefore by biometric process such fingerprint data. work [3], so in order to reduce such negativity related
Several advantages were found by using fingerprint. The traditional authentication systems biometrics was used.
simplicity of using fingerprint as a method to verify and prove Biometrics is the physiological and behavioural
the student identities helps allowing in entering the exam. In characteristics that can be measured in the human body and
other hand, the high level of accuracy offered by each single used to confirm the identity and differentiate it from others [4]
unique human fingerprint helps to prevent any method for
515 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
such as fingerprint. The strength of such biometric methods User could choose one of two categories of fingerprint-based
come from the inability to be stolen or lost as well as difficult system on the measurements. Unimodal systems uses only one
to be faked [3]. finger mostly the index (14), which proposed for low/medium
security places. Multi modal systems uses two or more fingers
Managing authentication is a very important activity by for authentication, mostly using the index and middle one,
fingerprints in order to ensure the integrity, accelerate the where it consider preferable for medium/high security items
process, decrease error rate and fasting verification process, [14]. These systems found to be best related to low error rates
where in a study reviled that time needs for students attendance and high efficiency.
verification using manual process 23.66 second per student,
more than 6.65 second using fingerprint [5]. A multi-intance fingerprint based authentication system has
been developed which consider more invincible to different
The most common mechanism for biometrics problems encountered in previous systems using the crossing
authentication systems consist of two phases [6]. The number technique [15]. The developed system considered
enrolment phase used in collecting data and mathematically highly efficient in verification the user with highly accuracy
analyzed it using specific algorithms to perform a data base. and low run time. Also, the system provides the flexibility to
The releases phase that interest in comparing and verifying the switch from multi-intance to unimodal in case of fault
identity. The general scenario for scanning the figure by using tolerance in order to preserve their independency [15].
sensor based on capturing two main fingerprint characteristics;
the valleys and ridges [7]. Indeed, the fingerprint based authentication system
developed to possesses the features of highly reliable and easy
Several reasons were considered as advantages for using for secondary development, as well as having several
fingerprint for authentication. The easy to apply is the most advantages such highly secure, highly accurate, easy in use and
important one; the low cost of the uses device as well as don’t being standardized make it applicable in different areas needs
need much power [6]. Although of the different advantages for authentication such as educational institutions, factories,
fingerprint some disadvantages presented with the complexity offices, security and access control systems [16].
in obtaining high quality images of finer patterns related to
present of tear, dirty and cuts of finger which will affect the III. DESIGN OF PROPOSED SYSTEM
accuracy, time of response and reliability [6].
The fingerprint-based system has been developed in the
In order to enhance fingerprint authentication system, context of examination system. Two phases have to be carried
different approaches have been deployed in authentication out; registration and verification.
system, where they in other hand have been fused in fingerprint
systems for indoor localization [8] by analyzing each indoor A. Registration Phase
algorithms to use the strengths and step down the weakness In this phase, the user (i.e. the student) is requested to add
points to build the best systems. the fingerprint in the database. The fingerprint is previously
processed to extract its features in which the fingerprints are
In other hand, effective identity authentication should be uniquely distinguished from each other. A website interface is
available for wireless networking. So a new robust firstly assigning the user’s Identifier UI (i.e. the university
authentication algorithm based on the phase noise fingerprint student number) to the scanned fingerprint. Once the
of the physical-layer was built [9]. As well as a security administrator send on order to scan the fingerprint, a secure
authentication scheme of combined physical-layers fingerprints connection is established between the front interface and the
to ensure the survivability of the network from attacks [9]. proposed device of the fingerprint-based system. In addition, a
The security associated with fingerprint-based system is an random password-based text is generated and assigned to the
important issue to be taken inconsideration. user identifier. The proposed system is then activated to start
the scanning process.
Fingerprint-based system has been addressed in different
applications. It was designed for ATM accesses, computer The user has to follow the prompt massages shown in the
network accesses, class room entering and building door looks LCD screen. The secure established connection which
[10]. Fingerprint authentication system presented also in automatically disconnected when the scanning process is
mobiles and smart devices where some security insights on activated. Another connection is established between the
touch dynamic provided [3]. New technique by using 3 proposed device and the target record at the database. Both the
dimensions magnetic finger motion pattern based implicit fingerprint features and the random password are integrated
authentication to provide highly accuracy was used in smart into the user’s record. The operation of the registration phase is
phones [11], [12]. given as follows:
For attendance checking, the used system built based on The user (i.e. student) Ui is given his identifier. The
fingerprint technology bonded with GPS presented in smart identifier represents the university student number STD.
phones to check user availability at anywhere [12]. In very
A random password is computed Ri = Rand(Ui ).
large countries like India, fingerprint-based system was used as
a voting system for all population by removing the Ui imprints his finger print FPion the sensor.
geographical constrains [13].
Compute IDi=h(FPi),where h(.) denotes one way hash
Using fingerprint as an access control in information function which is used to convert the FPi to identifier.
systems depends usually on user awareness and acceptance.
516 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
517 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
518 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
In comparison with other different devices, the proposed using artificial neural networks. International conference on computer,
system recognized itself from others by different communications and control technology (14CT), Kuching,1(1): 192-
195.
characteristics. Portability which means the capability to move
[5] I. A. Justina (2015). Fingerprint-based authentication system for time
the device from one place to another is applicable in the and attendance management. British journal of mathematics and
proposed system so it easy to use indoor and outdoor places not computer science, 5(6): 735-747.
like others used inside doors only [13],[17],[18]. [6] I. M. Alsaadi (2015). Physiological biometric authentication systems,
advantages, disadvantages and future development: a review.
The simplicity of the proposed device affects its ease of International journal of scientific and technology research, 4(12): 285-
development by different institutions where the approximate 289.
manufacturing cost around 100 dollars. This simplicity makes [7] A. Suganya, G. M. A. Sagayee (2015). A Delaunay pentangle-based
it cheaper as well as it is compatible with the android system fingerprint authentication system for preserving privacy using topology
"open source", on the contrary with other different systems code. International journal of research in engineering and advanced
considered expensive and mainly used paid operating system technology, 2(6): 142-149.
[18], [19]. [8] M. Chiputa, L. Xiangyang (2017). Real time Wi-Fi indoor positioning
system based on RSSI measurements: A distributed load approach with
Connections with external networks done by the control PC the fusion of three positioning algorithms. Wireless personal
(lap top) only which consider the first wall protection that communications: An international journal, 99(1): 67-83.
increase the security for the proposed device. In other hand, [9] C. Zhao, M. Huang, L.Huang, X. Du, M. Guizani (2017). A robust
authentication scheme based on physical-layer phase noise fingerprint
other devices depend on connecting the fingerprint device to for emerging wireless networks. Computer networks, 128(1): 164-171.
the external networks using network card, which could affect [10] D. Sunehra (2014). Fingerprint based biometric ATM authentication
the security and make them high risk to breakthrough [19],[20]. system. International journal of engineering inventions, 3(11): 22-28.
Furthermore the proposed device depends on a temporary [11] Y. Zhang, M. Yang, Z. Ling, Y. Liu, W. Wu (2018). Finger auth: 3D
communication channels with the device to send and receive magnetic finger motion pattern based implicit authentication for mobile
biometric data, after that the system disconnect immediately so devices. Future generation computer system (ISSN 0167-7399).
become harder to hack. [12] B. Soewito, F.L.Gaol, E. Simanjuntak, F. E. Gunawan (2015).
Attendance system on android smart phone. International conference on
VII. CONCLUSION control electronics, renewable energy and communications, Bandung,
pp: 208-211.
This research work discussed in detail fingerprint pre- [13] D. Khojare, V. Chaudhary, M. Malviya, Sh. Shukla (2018). FPKIVS-A
processing, minutiae extraction and minutiae matching. This stellar approach to voting systems in India. Advances in intelligent
research work has been able to provide a physical security and systems and computing, 653(1).
authentication for students before entering the class room. The [14] S. Ribaric and N. Pavesic (2008). A finger based identification system.
experimental result shows efficient registration and verification The 14th IEEE Mediterranean electrotechnical conference, Ajaccio, pp:
of subjects with accuracy over 98%. 816-821.
[15] A. Llugbusi and A. O. Adetunmbi (2017). Development of a multi-
ACKNOWLEDGMENT intance fingerprint based authentication system. International conference
on computing networking and informatics (ICCNI), Lagos, pp: 1-9.
Special thanks are given to our students Mohammad Abu [16] P. Sana, Sh. Prajakta, P. Kamini (2017). Fingerprint based exam hall
Osbah and Almuthna Alsaide for their efforts in items authentication system using microcontroller. International journal of
installation. Moreover, we would thank the president of Tafila engineering researches and management studies, 4(2): 89-91.
Technical University for his financial support. [17] B. Molina, E. Olivares, C. E. Palau, M. Esteve (2018). Amultimodal
fingerprint-based indoor positioning system for airports. IEEE Access,
REFERENCES 6(1): 10092-10106.
[1] J. Zhou, G. Su, Ch. Jiang, Y. Deng, C. Li (2007). A face and fingerprint [18] K. Chow, S. He, J. Tan, G. Chan (2019). Efficient locality classification
identity authentication system based on multi-rout detection. for indoor fingerprint based systems. IEEE Transactions on mobile
Neurocomputing, 70(1): 922-931. computing, 18(2): 290-304.
[2] M. H. Barkadehi, M. Nilashi, O. Ibrahim, A. Z. Fardi, S. Samad (2018). [19] J. J. Stephan, S. A. Abdullah, R. D. Resan (2017). Use fingerprint
Authentication systems: A literature review and classification. technology in developing country security. Annual conference on new
Telematics and Informatics, 35(5): 1491-1511. trends in information and communications technology applications,
[3] P. Sh. Teh, N. Zhang, A. B. J. Teoh, K. Chen (2016). A survey on touch Baghdad: 57-62.
dynamics authentication in mobile devices. Computer and Security, [20] J. Baidya, T. Saha, R. Moyashir, R. Palit (2017). Design and
59(1): 210-235. implementation of a fingerprint based lock system for shared access.
[4] M. H. Hammad, A. Mohammed, M. E. Eldow (2015). Design an IEEE 7th annual computing and communication workshop and
electronic system use the audio fingerprint to access virtual classroom conference, Las Vegas: 1-6.
519 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Cyber-security, as an emerging field of research, The level of research on the development of Intrusion
involves the development and management of techniques and Detection System (IDS) continues to increase as attacks
technologies for protection of data, information and devices. abound and attackers continue to evolve in practice. As a
Protection of network devices from attacks, threats and result, IDSs must evolve to prevail over the dynamic malicious
vulnerabilities both internally and externally had led to the activities carried out over a network. The development of a
development of ceaseless research into Network Intrusion Network Intrusion Detection System (NIDS) is critical for
Detection System (NIDS). Therefore, an empirical study was monitoring the network pattern behaviour of a computer
conducted on the effectiveness of deep learning and ensemble networked system [4]. Typically, an IDS monitors network
methods in NIDS, thereby contributing to knowledge by
packets to facilitate the identification of attacks and are
developing a NIDS through the implementation of machine and
deep-learning algorithms in various forms on recent network
basically categorised as either misuse/signature or anomaly
datasets that contains more recent attacks types and attackers’ based. Signature based IDS matches attacks to previously
behaviours (UNSW-NB15 dataset). This research involves the known attacks, and anomaly-based IDS uses the created
implementation of a deep-learning algorithm–Long Short-Term normal profile of a user to flag any profile that deviates from
Memory (LSTM)–and two ensemble methods (a homogeneous the user known behaviour [5].
method–using optimised bagged Random-Forest algorithm, and Because of the unrelenting efforts of attackers to
a heterogeneous method–an Averaged Probability method of
compromise a known network of computers and the new
Voting ensemble). The heterogeneous ensemble was based on
pattern of executing attacks and other malicious activities, the
four (4) standard classifiers with different computational
characteristics (Naïve Bayes, kNN, RIPPER and Decision Tree). need for a robust, up-to-date IDS is imminent to adequately
The respective model implementations were applied on the prevail against unknown attacks/threats or zero-day
UNSW_NB15 datasets in two forms: as a two-classed attack vulnerabilities.
dataset and as a multi-attack dataset. LSTM achieved a detection As such, an empirical research study was conducted to
accuracy rate of 80% on the two-classed attack dataset and 72% develop an IDS that can address new types of attacks in our
detection accuracy rate on the multi-attack dataset. The modern-day network using machine and deep learning
homogeneous method had an accuracy rate of 98% and 87.4%
algorithms. The contributions to knowledge produced during
on the two-class attack dataset and the multi-attack dataset,
respectively. Moreover, the heterogeneous model had 97% and
this research work are highlighted below:
85.23% detection accuracy rate on the two-class attack dataset 1) The use of more recent and complex network data as
and the multi-attack dataset, respectively. input data, i.e. the UNSW-NB15 dataset, for the development
Keywords—Cyber-security; intrusion detection system; deep of an IDS.
learning; ensemble methods; network attacks 2) Two (2) methods of implementing ensemble learning
methods for the development of an IDS;
I. INTRODUCTION 3) Implementation of a deep-learning technique (LSTM)
The proliferation of information and the technology used for building a NIDS;
for enabling communication in everyday life has prompted the 4) Development of two (2) categories of NIDS, i.e., two-
immense need for computer security [1]. The impact of class (normal and attack labels) and multi-attack (ten class
Information and Communication Technology on economic labels).
growth, social wellbeing, private and public business growth,
and national security is enormous as it provides the devices that Moreover, it is the intent of this research work to answer
propagate digital communications among hosts. The overall the following research questions:
protection of these hosts, which exist as computers, network 1) How effective is the ensemble learning method
devices, network infrastructures, etc. [2], as well as data and
implementation of NIDS for detecting attacks, both in a two-
information against cyber-attacks, worms, potential leakage
and information theft is fundamental to cyber-security [3]. class scenario and a multi-attack scenario?
520 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
2) How effective is the deep-learning implementation of recent dataset to uncover new malicious activities in a network
NIDS for detecting attacks, both in a two-class scenario and a [11]. Thus, UNSW-NB15 was used in this study. The UNSW-
multi-attack scenario? NB15 data was developed using the IXIA PerfectStorm tool in
3) What peculiarities are found in two-class and multi- the Cyber Range laboratory of the Australian Centre for Cyber
Security, which captured the sets of abnormal and modern-day
attack datasets and how do they affect the developed NIDS
normal network traffic. More details regarding the dataset
models? creation are given in [2].
II. RELATED WORKS Table I provides insights into the datasets used in this
The research conducted by [6] presented a deep-learning study.
method for developing a NIDS. The work proposed and As depicted in Table I above, the dataset is comprised of 45
implemented a Self-taught Learning (STL) deep-learning based attributes, of which, two (2) are dependent variables. Two
technique on a NSL-KDD dataset. The STL model when subsets of data are obtainable from the original dataset
evaluated based on training and test data achieved, in terms of according to the dependent variables; one of these subsets was
percentage, 88.39% accuracy for 2-class and 79.10% accuracy obtained to develop a two-class anomaly IDS, and the other
for 5-class. was use dot develop a multi-attack anomaly IDS. The
The work of [4] is a closely related work, wherein the distribution of the attacks is contained in the attack_cat
authors developed a multi-classification NIDS using the attribute, and the label attribute is comprised of normal and
UNSW-NB15 dataset and implemented an Online Average attack instances, denoted as 0 and 1, respectively.
One Dependence Estimator and an online Naïve Bayes with Regarding the features, Table II presents the details of both
83.47% and 69.60% accuracy, respectively. the independent and target variables.
Another research work conducted by [7] reported the use of Moreover, in light of data pre-processing and removal of
a deep neural network for development of a NIDS. The study redundant attributes, the first attribute indexed–id, serving as
implemented LSTM- Recurrent Neural Network (RNN) to the index of the dataset, was removed because it is irrelevant,
identify network behaviour as normal or affected based on the thus leaving two-class and multi-attack datasets with 43
past observations. KDDCup‟99 was used as the dataset, and the attributes each.
work achieved a maximum value of 93% efficiency.
Fig. 1 and Fig. 2 above depict the data distribution for both
The research work carried out by [8] developed four subsets of the original dataset. Fig. 1 depicts the ten (10) class
different IDS models using the RNN algorithm and tested them labels of the multi-attack dataset presented in Table I; each of
on a NSL-KDD dataset (binary and 5-classes) to evaluate the the labels is displayed using different colour. Fig. 2 shows the
models. The best model on a binary class achieved 98.1% two-class labels as presented in Table I above, with blue colour
accuracy using a 1-hidden layer BLSTM. For a 5-class, 87% representing the normal labels and red colour representing the
accuracy was achieved using a 1-hidden layer BLSTM. attack labels.
Using deep autoencoder (AE) after extracting features via
statistical analysis methods, [9] developed an IDS that TABLE. I. DATASET DESCRIPTION
achieved 87% accuracy on NSL-KDD dataset. Dataset Description
The study of [10] focused on using machine learning No. of Attributes 45
methods for developing an IDS using J48, MLP and Bayes No. of Independent Variables 43
Network (BN) algorithms to achieve the overall best accuracy No. Of Dependent Variables 2
of 93% with J48, 91.9% accuracy using MLP and accuracy of
Name: label
90.7% using BN on the KDD dataset. Details of the First Dependent
Normal Attack
Variable
III. METHOD 37,000 45,332
A. Dataset Details of the Second Dependent
Name: attack_cat
Variable
First, confirm that you have the correct template for your Normal 37,000
paper size. This template has been tailored for output on the
US-letter paper size. If you are using A4-sized paper, please Reconnaissance 3, 496
close this file and download the file “MSW_A4_format”. Backdoors 583
DoS 4,089
Most research studies on the development of IDS use the
KDDCup‟99 dataset; however, this dataset is gradually Exploits 11,132
becoming (if not already) obsolete because it does not contain Analysis 677
most new forms of attacks prevalent in modern networks of Fuzzers 6,062
computers. Reflection of contemporary threats and the Worms 44
inclusion of normal network packets are two important features
Shellcode 378
of a high-quality NIDS dataset. Because attackers execute
dynamic attacks daily, it is thus necessary to make use of a Generic 18,871
521 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
522 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. UNSW-NB15 ATTRIBUTES classification algorithms. The standard classifiers for the
heterogeneous ensemble considered in this study are described
No. Features No. Features
as follows: Bayes Theory (Naïve Bayes algorithm), Instance
1 id 23 dtrcpb Learning (k Nearest Neighbour), Rule-based (RIPPER) and
2 dur 24 dwin Tree methods (C4.5 Decision Tree). The voting combination
3 Proto 25 tcprtt method [14] [15] was adopted in this study for building the
4 Service 26 synack heterogeneous ensemble method. The voting method is a non-
complicated method of combining several predictions of varied
5 State 27 ackdat
or different models, and it can be implemented in a variety of
6 spkts 28 smean approaches, including majority vote, minority vote and average
7 dpkts 29 dmean of probabilities. The average of probabilities method of voting
8 sbytes 30 trans_depth [16] was selected for combining the results of each standard
9 dbytes 31 response_body_len classifier because the averaged results of the models are used to
10 rate 32 ct_srv_src
provide the final prediction.
11 sttl 33 ct_state_ttl DL is an advanced implementation of a neural network. A
12 dttl 34 ct_dst_ltm neural network is the simulation of the human brain, that is, a
13 sload 35 ct_src_dport_ltm
model of connected neurons. A neural network is usually
constructed to possess input, processing and output layers of
14 dload 36 ct_dst_sport_ltm
neurons [17]. The processing layer, often referred to as the
15 sloss 37 ct_dst_src_ltm hidden layer, may contain one or more layers–a basic
16 dloss 38 is_ftp_login implementation of neural network is the Multilayer Perceptron
17 sinpkt 39 ct_ftp_cmd (MLP) [18]. DL is an advancement on the MLP [19], but with
18 dinpkt 40 ct_flw_http_mthd more sophisticated and densely connected neurons that are
capable of representing and extracting data in a more advanced
19 sjit 41 ct_src_ltm
form from data and mapping it into the output [20, 21]. The
20 djit 42 ct_srv_dst neural network implementations that are used for DL include
21 swin 43 is_sm_ips_ports but not limited to Convolutional Neural Network, RNN and
22 stcpb 44 attack_cat Long Short-Term Memory (LSTM). In this study, the deep-
45 label learning method implemented was LSTM–a type of RNN. A
typical LSTM [7] consists of a cell, an input, an output, and a
B. Implemented Models forget gate, with which it captures the order dependence and
This empirical analysis implements three (3) different data recollection of values over random time intervals.
mining methods to develop a robust NIDS using both datasets Using the three (3) different data mining methods discussed
mentioned above. The approaches include: (i) Homogeneous above, several predictive models were developed using the
Ensemble, (ii) Heterogeneous ensemble, and (iii) Deep afore-mentioned datasets. Because it is known that model
Learning (DL) implementations. development is the next stage after the dataset and algorithm
An ensemble method [12] is the process of combining selection process and method identification phases, the
some different results, produced by contributing base learners, percentage split model development process was used in this
of predictive models via different combination methods to research work. The percentage split is the method of dividing a
make a final prediction based on aggregated learning. This given dataset into two: the first part is used for executing a
method is typically implemented via two phases: the first phase training phase-wherein the algorithms builds or fits their
being the construction of various models, and the second phase respective models, and the second part of the dataset is then
involving the combination of the estimates obtained from the used for testing–the phase whereby the fitted models are tested
various models [13]. The ensemble method is said to be by making predictions using the independent variables of the
homogeneous when the contributing base learners are multiples disjoint test set. Thus, a certain percentage value is given to
of the same computational characteristics (family). Base split the dataset into the training split and the test split.
learners in an ensemble model are standard classifiers. In this Moreover, having two datasets (two-class and multi-attack
study, the homogeneous ensemble was implemented in the datasets), each selected algorithm was fitted on each dataset
form of the Random-Forest (RF) algorithm. The Random- type, and the resulting models were tested on each
Forest algorithm is a bagging method that consists of a finite corresponding test sets, thereby producing some sets of models
number of decision tree algorithms with the addition of a that are categorised as (i) two-class attack anomaly IDS, and
„perturbation‟ of the classifier used for fitting the base learners. (ii) multi-attack anomaly IDS, each having three (3) separate
In particular, RF uses „subset splitting‟. The RF ensemble of models with respect to the applied method discussed above.
trees makes use of only a random subset of the variables while To summarise how the data mining methods were
building its trees; thus, the ensemble method is homogeneous. implemented and all robust NIDS models were all developed in
Alternatively, a heterogeneous ensemble is the combination this study, the proposed empirical framework is depicted in
of various results of base learners that have different learning Fig. 3, and the experimental results produced are presented in
methods or computational characteristics, that is, the table and charts and extensively discussed as seen in sections
contributing base learners belong to different categories of below.
523 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
C. Performance Evalutaion Metrics using various metrics as appropriate for the type of the
Following the model development process stage, the developed IDS model.
developed models are evaluated. As such, the performances of Conclusively, all experiments were carried using Waikato
models were evaluated based on the category they belonged to. Environment for Knowledge Analysis (WEKA) tool for data
The two-class anomaly IDS models were evaluated using the analysis, wherein results were all obtained and presented in
following metrics [17]: Detection rate, Area Under Curve relevant section of this paper.
(AUC), True Positive (TP), False Positive (FP), True Negative
(TN) and False Negative (FN). The multi-attack anomaly IDS TABLE. III. IMPLEMENTATION OF EACH ALGORITHMS
models were evaluated based on the following metrics [18]:
Detection rate, Kappa value and Weighted (AUC, TP, FP and Algorithm Parameter Settings
F-measure). The multi-attack models were evaluated using NB
useKernelEstimator = True; useSupervisedDiscretisation =
weighted values because of the multiple values of the class False, batchsize = 100
labels (ten in number), unlike the two-class anomaly IDS, kNN windowSize = 0, batchsize = 100
which has just two classes (normal and attack)–a binary RIPPER
usePruning = True, seed = 1, batchsize = 100; folds = 5,
classification model. minNo = 2.0; optimisations = 2, checkErrorRate = True
batchsize = 100, binarySplits = False collapseTree = True;
The proposed empirical framework presented in Fig. 3 C4.5 Decision confidenceFactor = 0.25; minNumObj = 2; numFolds = 5;
above consists of the Data Pre-Processing and Re-Labelling Tree subtreeRaising = True, unpruned = False; seed = 1.
Module and the Method Module, which interacts with the .useLaplace = False; useMDLcorrection = True
bagSizePercent = 100; batchSize = 100;
Model Development Process Module in producing the two breakTiesRandomly = False; maxDepth = 0;
forms of IDS mentioned in this study. The Algorithm module RF
computeAtrributeImportance = False; numFeatures = 30;
consists of the selected algorithms for this study, and this numIterations = 20; seed = 1
module interacts with the Method module, which defined the reluAlpha = 0.01, Updater = adam, OptimizationAlgorithm
data mining implementations. Last, the Metrics component = SGD, learning rate = 0.1, dataset= standardise. While
evaluates the produced model based on its form, and the developing the two-class anomaly IDS, LSTM layer was
LSTM (two-
configured as neurons = 128, activation function = ReLU,
evaluation results are subsequently discussed. class)
gate activation function= Sigmoid, dropout = 0.3; Output
layer parameter was lossFunction = LossMCXENT,
Table III presents the parameter settings for each algorithm
activation function = softmax
used in this study. All models were trained and tested using the LSTM (Multi- activation function = softmax; gate activation function =
percentage split strategy–80% was used for training and 20% attack) ReLU
was used for testing, and their performances were evaluated
Methods Algorithms
UNSW_NB15
Heterogeneous Ensemble
Standard Classifiers
Random Forest
Decision Tree
Deep Learning
Model Dev. Process
524 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
IV. RESULTS From Table VI, the heterogeneous ensemble had an overall
Having implemented the proposed framework of this detection rate of 96.92% with an AUC score of 0.994,
research, the reported results will be categorised into two indicating yet another very strong prediction model. The TP
according to the model development processes. Note that the value of 0.98 indicates that the model classified 98% of normal
test was conducted on 20% of the dataset, resulting in 16,466 packets as normal, and the TN value of 0.957 denotes that the
instances. First, the two-class anomaly IDS is basically the attack packets were correctly classed as attack at the rate of
prediction of whether a network packet is normal or an attack 95.7%. The FP value of 0.43 denotes that approximately 5% of
and is thus evaluated using the given metrics in Fig. 3. For the normal packets were classified as attack, and the FN value of
homogeneous method, Tables IV and V present the 0.016 indicates that approximately 1.6% of attack packets were
performance scores of the model and its corresponding predicted as normal. Likewise, Table VII–the confusion matrix
confusion matrix, respectively. of the heterogeneous ensemble, depicts the actual figures of the
TP–7277 of 7395 normal instances classified as normal, FP–390
From Table IV, the homogeneous ensemble had an overall of 9071 attack instances were misclassified as normal instances,
detection rate of 97.96% with an AUC score of 0.997, TN–8681 of 9071 attack instances correctly classified as attack
indicating a very strong prediction model. The TP value of and FN–118 of 7395 normal instances misclassified as attack.
0.98 indicates that the model classified 98% of normal packets
Last in this category, the results of deep-learning method
as normal, and the TN value of 0.976 denotes that the attack
for developing a two-class anomaly IDS as implemented with
packets were correctly classed as attack at the rate of 97.6%.
the specified parameters described in the previous section are
The FP value of 0.024 denotes that just 2% of normal packets
shown in Tables VIII and IX.
were classified as attack, and the FN value of 0.0158 indicates
that approximately 1.58% of attack packets were predicted as Table VIII shows that the deep leaning model had an
normal. Likewise, Table V–the confusion matrix of the overall detection rate of 80.72% with an AUC score of 0.926,
homogeneous ensemble, depicts the actual figures of the TP– i.e. the deep-learning model is a competitive predictive model.
7278 of 7395 normal instances classified as normal, FP–219 of The TP value of 0.57 indicates that the model classified 57% of
9071 attack instances misclassified as normal, TN–8852 of normal packets as normal–a fair result as compared to other
9071 attack instances correctly classified as attack, and FN– models in this category; it has a strong TN value of 0.998,
117 of normal instances misclassified as attack. indicating that the attack packets were correctly classed as
attack at the rate of 99.8%-the best TN value in this category.
For the heterogeneous ensemble, the voting cum average
The model had a FP value of 0.002, denoting an insignificant
probabilities results for different techniques are shown in
number of misclassified normal instances, and the FN value of
Tables VI and VII below.
0.426 indicates that approximately 42.6% of attack packets
TABLE. IV. HOMOGENEOUS ENSEMBLE MODEL‟S EVALUATION were predicted as normal. Likewise, Table IX – the confusion
matrix of the heterogeneous ensemble, depicts the actual
Evaluation Metric Score figures of the TP–4239 of 7395 normal instances classified as
AUC 0.997 normal, FP–19 of 9071 attack instances misclassified as
normal, TN–9052 of 9071 attack instances correctly classified
TP Rate 0.984178
as attack and FN–3156 of normal instances misclassified as attack.
FP Rate 0.024143
TN Rate 0.975857 TABLE. VII. HETEROGENEOUS MODEL CONFUSION MATRIX
525 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Critical evaluation of the models in this category reveals TABLE. XII. DEEP-LEARNING MODEL‟S EVALUATION OF MULTI-ATTACK
ANOMALY IDS
that, despite all models performing well using the AUC metric,
the deep-learning model is weak in the detection of normal Evaluation Metric Score
packets and will generate more false flagging of normal
Weighted AUC 0.868
packets, thereby degrading the network monitoring in real
time. Moreover, although the homogeneous and heterogeneous Weighted TP Rate 0.723
models competed fairly with each other, as they are both robust Weighted FP Rate 0.171
models for the detection of normal and attack packets, the
Weighted F-Measure 0.659
homogeneous ensemble model is the best model in terms of
lower FP and higher AUC values. Kappa Statistics 0.57
The second category is the multi-attack anomaly IDS, Detection rate 72.26
which is the classification of packets into normal and nine
In this multi-attack category, the homogeneous ensemble
different types of attacks–a typical multi-classification
method also achieved the best performance, with a weighted F-
problem, as discussed in previous section. The models are
measure of 0.87, a kappa value of 0.82, and an overall
evaluated as depicted in Fig. 3. For the homogeneous ensemble
detection rate of 87%. Although the heterogeneous had a
method in this category, Table X reveals various performances
weighted AUC of 0.982, it is the second best in this category.
scores.
Last, the deep-learning model competed fairly well with the
Table X reveals the model‟s ability to detect whether a other models, with its weighted AUC of 0.868; however, it had
packet belongs to any of the ten (10) classes at 87.39%. This a low kappa value of 0.57 and a low detection rate of 72.26.
model had a kappa value of 0.8 and a weighted AUC of 0.98. Moreover, the confusion matrix for each model reveals the
The weighted TP value is 87.4%, and the weighted FP value is classification and misclassification of the instances
2.5%. The model also had a weighted F-measure value of 0.87. accordingly. The deep-learning model was found to be unable
to detect many attack classes, whereas the homogeneous model
Similarly, in Table XI, this model detection rate was was adequately robust.
85.23% but with a weighted AUC of 0.98, a weighted TP value
of 0.852–85.2% correct classification of each class label A summary of the detection rate of all models for both
instances, a weighted FP value of 0.031, a weighted F-measure categories is presented in Table XIII.
of 0.855, and a kappa value of 0.79.
Table XIII concisely presents the detection rates for all the
Last in this category, the deep-learning model of multi- above-described models, as is pictorially depicted in Fig. 4.
attack anomaly IDS was also evaluated; its scores are
represented in Table XII. TABLE. XIII. SUMMARY OF THE RESULTS
The deep-learning model yielded an ability to detect and Models Methods Detection Rate (%)
predict the class of any packet at 72%. This result is achieved Homogeneous Ensemble 97.96
at a weighted AUC value of 0.868, a weighted F-measure score
Two-class Anomaly IDS Heterogeneous Ensemble 96.92
of 0.659, and a kappa value of 0.57. This model is capable of
correctly detecting each class instance at the weighted TP value Deep Learning 80.72
of 72.3, and it had a weighted FP value of 0.17. Homogeneous Ensemble 87.39
Multi-attack Anomaly
TABLE. X. HOMOGENEOUS MODEL‟S EVALUATION Heterogeneous Ensemble 85.23
IDS
Deep Learning 72.26
Evaluation Metric Score
Weighted AUC 0.98
Weighted TP Rate 0.874 Detection Rate
Weighted FP Rate 0.025
Homo. Hetero. DL
97.96
96.92
85.23
526 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
527 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[4] M. Nawir, A. Amir, N. Yaakob, and O. N. G. B. I. Lynn, “Multi- [14] M. Sabzevari and G. Mart, “Vote-boosting ensembles,” Pattern
Classification of Unsw-Nb15 Dataset for,” vol. 96, no. 15, pp. 5094– Recognit., 2018.
5104, 2018. [15] D. Murphree et al., “Ensemble Learning Approaches to Predicting
[5] S. M. Thaler, Automation for information security using machine Complications of Blood Transfusion,” in IEEE Eng Med Biol Soc.,
learning. 2019. 2016, pp. 1–11.
[6] Q. Niyaz, W. Sun, A. Y. Javaid, and M. Alam, “A Deep Learning [16] I. H. Witten, E. Frank, and M. A. Hall, Data Mining - Practical Machine
Approach for Network Intrusion Detection System,” 2016. Learning Tools and Techniques. 2011.
[7] M. Ponkarthika and V. R. Saraswathy, “Network Intrusion Detection [17] Y. Xin et al., “Machine Learning and Deep Learning Methods for
Using Deep Neural Networks,” vol. 2, no. 2, pp. 665–673, 2018. Cybersecurity,” IEEE Access, vol. 6, pp. 35365–35381, 2018.
[8] A. Elsherif, “Automatic Intrusion Detection System Using Deep [18] M. A. Mabayoje, A. O. Balogun, A. O. Ameen, and V. E. Adeyemo,
Recurrent Neural Network Paradigm,” 2018. “Influence of Feature Selection on Multi - Layer Perceptron Classifier
[9] C. Ieracitano, A. Adeel, M. Gogate, K. Dashtipour, and C. R. Aug, for Intrusion Detection System,” Comput. Inf. Syst. Dev. Informatics
“Statistical Analysis Driven Optimized Deep,” Int. Conf. Brain Inspired Allied Res. J., vol. 7, no. 4, pp. 87–94, 2016.
Cogn. Syst. Springer, Cham., pp. 759–769, 2018. [19] F. Feng, X. Liu, B. Yong, R. Zhou, and Q. Zhou, “Anomaly detection in
[10] M. Alkasassbeh and M. Almseidin, “Machine Learning Methods for ad-hoc networks based on deep learning model: A plug and play
Network Intrusion Detection,” no. October, 2018. device,” Ad Hoc Networks, 2018.
[11] G. Li and Z. Yan, “Data Fusion for Network Intrusion Detection : A [20] D. Papamartzivanos and G. Kambourakis, “Introducing Deep Learning
Review,” vol. 2018, 2018. Self-Adaptive Misuse Network Intrusion Detection Systems,” IEEE
Access, vol. 7, 2019.
[12] G. Seni and J. F. Elder, Ensemble Methods in Data Mining: Improving
Accuracy Through Combining Predictions. 2010. [21] SH Kok, Azween Abdullah, NZ Jhanjhi, Mahadevan Supramaniam, “A
Review of Intrusion Detection System Using Machine Learning
[13] P. Illy, G. Kaddoum, C. M. Moreira, K. Kaur, and S. Garg, “Securing Approach”, in International Journal of Engineering and Research, Jan
Fog-to-Things Environment Using Intrusion Detection System Based 2019.
On Ensemble Learning,” no. April, pp. 15–18, 2019.
528 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Awais Qasim1 , Sidra Kanwal2 , Adnan Khalid3 , Syed Asad Raza Kazmi4 , Jawad Hassan5
Department of Computer Science, Government College University, Lahore Pakistan
Abstract—This research focuses on Timed-Arc Petri-nets- agents and portray how collaborative agents response on mes-
based agent communication in real-time multi-agent systems. sages. It is normal to make models from basic conventions
The Agent Communication Language is a standard language for into complicated protocol. These interaction protocols have
the agents to communicate. The objective is to combine Timed- been demonstrated in MAS but not presented in RTMAS. FIPA
Arc Petri-nets and FIPA Performatives in real-time multi-agent performatives provide an outline for the existence and action of
systems. FIPA standards provide a richer framework for the
agents. Existing FIPA protocols have been described through
interaction of agents and makes it easier to develop a well-defined
system. It also ensures the management by precisely specifying Petri-nets but the timing aspect was not specified which was
the agent’s interaction. Though FIPA protocol has already been a limitation. The main idea of this research is to formally
described with the help of Petri-nets but this specification lacks model the existing FIPA performatives by Timed Arc Petri-
the timing aspect that is a dire need for real-time multi-agent Nets (TAPNs) in real-time multi-agent systems. RTMAS is
systems. The main objective of this research is to provide a formally determined and checked with the time limitations.
method of modeling existing FIPA performatives by combining In a real time system, there are some actions, which have
Timed-Arc Petri-nets in real-time multi-agent systems. We have specified deadlines and depicts how long agents will wait for
used properties, such as liveness, deadlock and reachability for reply or perform next action by using FIPA performatives.
the formal verification of the proposed modeling technique.
Keywords—Formal verification; FIPA; multi-agent systems; Agent’s communication has been modeled formally in the
timed-arc petri nets; real-time systems past but not for real-time environment. The communication
between agents is modeled by using Agent Petri-nets (APN)
I. I NTRODUCTION and it is undeniable that integration between protocols and
APN greatly facilitates the development of a system which
Modeling of agents for the transmission of messages is leads to correct interaction between agents through appro-
much needed as agents interact with one another to achieve priate specification of the exchange of messages. The time
goals. Cooperation in Multi-Agent Systems (MAS) is mainly aspect has not been handled in [2] before performing any
achieved through interacting agents. For effective communica- target. The work of [3], [4], [5] leads to formal specification
tion these interacting agents require some interaction protocol. and verification of interactive real-time software agents (RT
The main purpose of the interaction protocol is to provide Agents). Agents work independently and handle the uncertain
a set of well defined rules for the communication of agents. scenarios. Visually expressive broader structure and modeling
The primary function of an agent is to handle the dynamic approach i.e. TAPN have been used for specification and
situations. There is no agent that possesses information of the representation of Stock Market System (SMS). It is based
whole system rather its decision making is dependent upon on RTMAS. The Model is verified by Timed Computational
limited view of the complete system [1]. Real-time agents can Tree Logic (TCTL) fragments AF, AG, EG and EF. In this
be depicted within the deadline. MAS’s have been formally paper, KQML register conversation and simple negotiation
specified using petri-nets but no such work has been done interaction conversation are modeled through CPN. The work
for the formal modeling of agent communication protocols done in [6] describes popular ACL like FIPA to formally model
in real-time environment. By using the Real-Time Multi- the organization of MASs and clarifies the analysis about
Agent Systems (RTMAS), the interaction of agents is bounded FIPA ACL semantics. FIPA specification has been used for
within the proposed context. Protocols become the cause of guidance. ACL specifications have been introduced with the
message flow among these RTMAS’s and specify the sequence example of online stock brokering to secure the interaction
of messages, number of messages and updation. Foundation between agents. This new model of ACL has expressiveness
for Intelligent Physical Agent (FIPA) performatives provide and reusability. In [7] nested petri-nets have been used to
outline for the existence of agents, their actions and archi- model multi-agent systems. In [8] the overview of FIPA-ACL
tecture. They also elaborate the authentication of the agents. and protocols has been given. FIPA-ACL is based on speech
Agent Communication Language (ACL) is a proposed standard act theory as communicative act. With the passage of time,
language for agent communications like FIPA-ACL. FIPA has a lot of improvements have been made but still, none of the
20 performatives i-e request, inform, accept proposal, etc that protocols can be treated as the complete in itself. In [9] an
describe the interpretation of messages. These messages are ac- agent-based framework in an unconstrained platform has been
tually the actions (communicative actions or CAs). Interaction described. This research highlights agents and multi-agents
protocols are the standards that oversee the interaction between system as the state-of-the-art and distributed environments
agents. It permits the description of expressly sequence of di- respectively. The work of [10] emphasizes the modeling of the
alogue between agent’s communication. Interaction Protocols vehicle framework that can go through crossing points with
are utilized to characterize set of messages transmitted between no or less delay. This approach determines how the model
www.ijacsa.thesai.org 529 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
emphasizes the traffic movement (transitions) by reachability formally specified in TAPN. TAPN is a framework for visual
graph within time constraints. Time-based constraints for MAS representation of sequence of events in time. This is used to
are presented in [11]. It provides clear ways to accomplish describe the modeling approach. Arcs are used to represent the
the MAS compliance. Vehicle to vehicle communication is time specification i-e inhibitor arc, transport arc and invariants.
presented by using Petri-nets. Vehicle to vehicle commu- In the formal specification of RTMAS, agents interact with
nication makes an efficient exchange of messages between one another to achieve their goals within time constraints.
cars and also matches the ID of cars for verification. It This element is required for their correct functioning. Protocols
also works for the modeling and stimulation of vehicles to define how much long period of time the agents would wait
vehicle communication of discrete event system, Petri-net is for the concerned interacting agents and also for updation.
used as a powerful tool. In [12] a more far-reaching display TAPAAL model checker is used for formal verification of
and re-enactment approach is introduced that records for the FIPA performatives in RTMASs. TAPAAL is a graphical
MAS-related conventions as portrayed in the FIPA particular; representation and verification tool of TAPN. It is also used
together with a co-reproduction stage for the examination of for the verification of different queries specified to ensure
MAS. The process of checking a smart workflow management the correct functioning of the system’s model. The model is
framework and postpone forecast is presented [13], [14]. The verified with the fragments of TCTL whose fragments are AF,
work of [15] describes the formal specification approach for AG, EG and EF.
the presentation of communicative agents. It explores the
internal state of agents and behavior of interactive agents. II. P RELIMINARIES
In [16] model checking techniques are discussed, which are
utilized as a part of tool TAPAAL for reachability analysis A few terminologies and computational models have been
using inhibitor arcs. Reachability is utilized for the dead state. described in this section that will be used in the rest of the
It finds that there is a state which is not reachable from some discussion for the specification of the problem under analysis.
other state. The research of [17] presents a liveness based
analysis for Timed-Arc Petri-Nets with weights and arcs i.e. A. Timed-Arc Petri-Net (TAPN)
inhibitor arcs, transport arcs and age invariants. This research TAPN is an established technique for the formal modeling
highlights agents and MAS as state of the art and distributed of multi-agents systems in real-time environments. The timing
environments, respectively. Face detector and tracker agents aspect considers unequivocal treatment of real-time. TAPN
interact through contract net protocol. This system tracks the provides a mechanism to formally verify the properties of
agents which path has been chosen and time spent. Events interest of the system to ensure its correctness. The TAPN
triggered sample data consensus has been proposed in [18], definition has been proposed in [24]. A TAPN is a 6-tuple (P,
[19]. T, IA, OA, I, Type) where
Even triggering condition is intermittently examined after
P: is finite set of places.
constant sampling instant for distributed MAS with directed
graph. Consensus of distributed MAS can be transformed T: is a finite set of transitions.
into stability of system with time. Then a sufficient condition
on the consensus of Multi-Agent system is derived. The IA: P*T is finite set of input arcs.
management of communication between agents is presented OA: T*P is finite set of output arcs.
in [20]. This model is illustrated with the help of Petri-nets
and result is validated with coordination between agents. In I: represents the age interval of places.
[21] demonstrates that response to the presence of discrete- Type represents the type of arc normal, inhibitor, transport
time and continuous time strategies occur at the same time arc.
or getting input or wait with decision until next occasion is
activated and creates algorithm for discrete Timed-arc Petri- TAPNs are basically an extension of the standard petri-nets
net games. A work process based that concentrates on the in which age of the tokens is utilized to incorporate timing
foundational issues of soundness is strongly based on timed- aspect of the system. The time intervals defined on arcs are
constrained soundness. Through subclass of bounded nets, used to restrict the progress of the system by only allowing
we can efficiently verify the design [22]. For the effective those tokens whose age falls in the interval. Arc defines a
processing, agents search for more agents by using KQML. place to transition and transition to place. Transition firing and
Contract net protocol is utilized on the interaction of agents. enabling depend on the age of token. Transition enabling and
If an agents bounds with one contract, it is illegible to take firing is not possible when the time interval is expired that
new one until the completion of previous [23]. is mentioned on a certain arc. There are three types of arcs
such as transport arc, inhibitor arc and normal arc. Transport
According to our knowledge, formal specification of arc produces the same time which is consumed on input arc.
MAS’s interaction in a real-time environment is a novel Inhibitor arc restricts the certain age of token on the place.
approach. Although formal modeling of MASs have been done Normal arc produces the age zero token although consumed of
in the past but it is limited to domain functionality of the any age token. TAPN is used for the verification of reachability,
complete system either at the micro or at macro level. In boundedness and liveness properties.
MAS goals can be defined at any single agent level or at
the complete system level but any single goal may contain B. TAPAAL
several agent performatives. Formal modeling of agents inter-
action allows the designer to verify the system’s correctness TAPAAL is the graphical representation and powerful tool
at the design time. In this research, FIPA performatives are for modeling of Timed-Arc Petri-nets. TAPAAL is used for
www.ijacsa.thesai.org 530 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
simulation and verification of TAPN. It is graphical, modeling and <Request A1 A2> to A2. Time interval on arc re-
editor for Timed-Arc Petri-net. It provides its own engine stricts the token firing time that must be fired within [StRe-
for verification [25]. For verification different properties are quest, EnRequest] including delay. Agent2 accepts the re-
used such as Reachability, boundedness, and liveness. These quest by passing the message of Agree as transition firing
properties are defined under the Timed Computational Tree <Agree A2 A1> and can also refuse the request as transition
Logic (TCTL). TCTL is the actual logic which is utilized for firing <Refuse A2 A1>. A1 receives the response of Agent2.
the determination of properties about structure. The fragments In case Agent2 agrees to the request of Agent1, token is fired
of TCTL are AF, AG, FG, and EF. on Agent1 place with the age 0 for the response of agree and
another token is fired on Agent2 place to continue the process
C. FIPA Performatives with same age that has been consumed at the fired transition.
For this purpose, we have used transport arc.
FIPA developed its standards in 1995 for agent’s commu-
nication. Speech act theory provides a base for FIPA ACL.
Modeling of agents is useful for the transmission of messages.
Agents interact with one another for the achievement of some
goals. Agent communication language is a proposed standard
language for agent communications like FIPA-ACL. In [26]
FIPA provides a specification for agent Communication. FIPA-
ACL suggests the parameters for effective communication
among the agents within the scenario. All performatives of
FIPA contain message structure i-e sender, receiver and content
of the message. The message expresses the meaning of agents.
It consists of the content and action of the communication.
If an agent communicates with another agent, a suitable
performative is used. If the agent does not understand some
message and is unable to the process the sent message, it can
reply with the “not understand” performative. In [26] FIPA has
proposed 20 performatives that cover all the maximum range of
aspect of agents expected communication. FIPA-ACL is now
a basic standard for determining the encoding and exchange
of messages among agents. In [27] a set of performatives is
given for ACLs that specifies how these communicating actions
should be executed in a concurrent and reactive way with
respect to a given logical semantics. It has assumptions about
how the recipient should react to the message.
TABLE I. T RANSITION AND TIMING DURATION OF TWO AGENTS IN CFP performative, one agent acts as a manager, that requires a
R EQUEST P ERFORMATIVE . certain task to be accomplished effectively within a specified
time. In CFP performative, the manager sends the call for
Place Transition Total Time Duration Start Time End Time proposal to the contractors. The contractors send the reply
Agent1 Request A1 A2 7 0 7 in one of the three forms that are refused, not understand-
Agent2 Refuse A2 A1 7 8 15 ing and proposal within the time limit. After getting the
Agent2 Agree A2 A1 7 8 15
Agent2 InformDone A2 A1 7 16 23 response from the contractors, the manager approves one of
Agent2 InfoRef A2 A1 7 16 23 the proposals and sends rejection (reject proposal) to the
Agent2 Failure A2 A1 7 16 23 remaining agents. The selected contractor then apprises the
manager of task completion. In CFP performative, we have
used three places and fifteen transitions. Agent1 place has
TABLE II. T RANSITION AND TIMING DURATION OF ONE TO MANY 0.0 token that represents the age of the token. Agent1 CFP
AGENTS INTERACTION IN CFP P ERFORMATIVE . to Agent2 and Agent3 is represented as transition firing
<CFP A1 A2A3>. Time interval on arc restricts the token
Place Transition Total Time Duration Start Time End Time firing time that must be fired within [StCFP, EnCFP] including
Agent1 CFP A1 A2A3 7 0 7 delay. Agent2 accepts the message by passing the proposal
Agent2 Proposal A2 A1 7 8 15 as transition firing <Proposal A2 A1> and refuses transition
Agent2 Refuse A2 A1 7 8 15
Agent2 NotUnderstand A2 A1 7 8 15 firing <Refuse A2 A1>. Agent1 can accept the proposal as
Agent3 Proposal A3 A1 7 8 15 transition firing <AcceptProposal A1 A2> within [StAccept
Agent3 Refuse A3 A1 7 8 15
Agent3 NotUnderstand A3 A1 7 8 15
, EnAccept]. For accept Proposal, we have used transport
Agent1 RejectProposal A1 A2 7 16 23 arc which means that the age of the token remains same
Agent1 RejectProposal A1 A2 7 16 23 as the age of token at the firing time. Agent1 can reject
Agent1 AcceptProposal A1 A2 7 24 31
Agent1 AcceptProposal A1 A3 7 24 31 the proposal as transition firing <RejectProposal A1 A2>
Agent2 InfoDone A2 A1 7 32 39 with time constraints [StReject, EnReject]. In case of re-
Agent2 InfoRef A2 A1 7 32 39 ject proposal, the token is transferred to Agent2 or Agent3.
Agent2 Failure A2 A1 7 32 39
Agent3 InfoDone A3 A1 7 32 39 The agent replies further within time constraint if the pro-
Agent3 InfoRef A3 A1 7 32 39 posal is accepted by Agent1. After Acceptance of proposal,
Agent3 Failure A3 A1 7 32 39
Agent2 replies in the form of <InformDone A2 A1> or
<InformDone A3 A1> within [StInfDone, EnInfDone] if the
task is completed within deadline. Agent2 or Agent3 can reply
B. Call for Proposal in the form of <InfoRef A2 A1> <InforRef A3 A1> within
Call for proposal (CFP) is used to start communication [StInfoRef, EnInfoRef] if detailed reply. If agent fails to fulfil
between agents. FIPA CFP performative augments approval the proposal requirements within fixed deadline then it replies
and denial of preceding form of communicating agents. In as <Failure A2 A1> or <Failure A3 A1> within [StFailure,
www.ijacsa.thesai.org 532 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 2. TAPN based CFP Protocol for one to many agents interaction.
TABLE III. P ROPERTIES FOR R EQUEST P ERFORMATIVE AND TCTL TABLE VII. K- BOUNDEDNESS OF CFP P ERFORMATIVE “A LL THESE
FRAGMENTS FOR TAPAAL. STATES REACHABLE ?”
www.ijacsa.thesai.org 534 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE X. E XECUTION C OUNT FOR ALL THE S IGNALS IN TAPAAL. TABLE XI. E XECUTION C OUNT FOR T RAFFIC L IGHT S YSTEMS IN
TAPAAL.
Count Transition
Count Transition
31 TLS.Acc Op Sig1
31 TLS.Acc Op Sig4 1122 TLS.Con Per ActG1
31 TLS.Acc Op Sig3 967 TLS.Con Per ActG3
31 TLS.Acc Op Sig2 891 TLS.Con Per ActG4
26 TLS.Rej St Sig2 846 TLS.Con Per ActG2
26 TLS.Rej St Sig1 725 TLS.Acc Op Sig1
26 TLS.Rej St Sig 725 TLS.Acc Op Sig4
26 TLS.Rej St Sig3 725 TLS.Acc Op Sig3
21 TLS.Inf Per Act1 725 TLS.Acc Op Sig2
21 TLS.Inf Per Act3 673 TLS.Go1
21 TLS.Inf Per Act4 528 TLS.Go3
21 TLS.Inf Per Act2 465 TLS.Go4
14 TLS.Con Per ActG1 431 TLS.Go2
13 TLS.Con Per ActR1 421 TLS.Con Per ActR1
12 TLS.Pro N Obj1 388 TLS.Con Per ActR3
12 TLS.InfR N Obj1 372 TLS.Con Per ActR4
8 TLS.Pro N Sig3 367 TLS.Stop1
8 TLS.Con Per ActG3 364 TLS.Con Per ActR
8 TLS.InfR N Obj3 299 TLS.Stop3
8 TLS.Con Per ActR3 272 TLS.Stop4
7 TLS.Pro N Obj4 256 TLS.Stop2
7 TLS.Con Per ActR4 165 TLS.Rej St Sig2
7 TLS.Con Per ActG4 165 TLS.Rej St Sig1
7 TLS.InfR N Obj4 165 TLS.Rej St Sig
6 TLS.Pro N Obj2 165 TLS.Rej ST Sig3
6 TLS.InfR N Obj2 159 TLS.Pro N Obj1
6 TLS.Con Per ActG2 159 TLS.InfR N Obj1
6 TLS.Con Per ActR2 154 TLS.Pro N Sig3
2 TLS.Go1 154 TLS.InfR N Obj3
2 TLS.Stop1 151 TLS.Pro N Obj4
2 TLS.Inf N Obj1 151 TLS.InfR N Obj4
1 TLS.Inf N Obj2 150 TLS.Pro N Obj2
1 TLS.Go2 150 TLS.InfR N Obj2
1 TLS.Stop2 135 TLS.Inf N Obj1
1 TLS.Stop4 122 TLS.Inf N Obj3
1 TLS.Go4 118 TLS.Inf N Obj4
1 TLS.Inf N Obj4 115 TLS.Inf N Obj2
1 TLS.Go3 35 TLS.Inf Per Act1
1 TLS.Inf N Obj3 35 TLS.Inf Per Act3
1 TLS.Stop3 35 TLS.Inf Per Act4
35 TLS.Inf Per Act2
senses the presence of vehicles and gives information to the tance message to the system that consists of the maximum
system using <Inf N Obj> transition that is completed in number of vehicles and the System sends a message to
[StInfO,EnInfO]. Now each system tells the controller about <YG Traffic Light> to get ready. Then <YG Traffic Light>
the number of vehicles using the proposal interaction pro- is shifted to <G Traffic light> using go transition within
tocol. Transition varies from system to <Pro N Obj> and [StG, EnG]. In Table IX, verification results of the properties
<Pro N Obj> to the controller. The controller checks each of interest of the TLS specified in TCTL fragments are shown.
system proposal and finds the maximum number of vehicles. It Table X and Table XI shows the transition count of different
rejects the proposal of system that does not have the maximum arcs during simulation. The higher number of counts represents
number of vehicles. Next, the system sends message to yellow more traffic at that signal.
traffic light agent that is <YR Traffic Light> to get ready
and the <YR Traffic Light> is turned to <R Traffic light> TAPN has been used for the verification of boundedness,
to perform the action of stop. The controller sends the accep- reachability and liveness properties. Formal verification gives
www.ijacsa.thesai.org 536 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
the correctness of the system. Boundedness ensures the max- [9] C. Zaghetto, L. H. M. Aguiar, A. Zaghetto, C. G. Ralha, and F. de Bar-
imum and the minimum number of tokens that each place ros Vidal, “Agent-based framework to individual tracking in uncon-
holds. Reachability determines the sequence from the first node strained environments,” Expert Systems with Applications, vol. 87, pp.
118–128, 2017.
of marking to the second node. In our proposed application, [10] Y.-S. Huang, Y.-S. Weng, and M. Zhou, “Design of traffic safety control
all the states are reachable. Liveness determines that the systems for emergency vehicle preemption using timed petri nets,” IEEE
application is executable. All places can contain any number of Transactions on Intelligent Transportation Systems, vol. 16, no. 4, pp.
tokens throughout the life cycle. Our system is deadlock free 2113–2120, 2015.
because there is no such place where deadlock can occur. The [11] D. Calvaresi, M. Marinoni, A. Sturm, M. Schumacher, and G. Buttazzo,
dead-lock free system implies that the system is live and all “The challenge of real-time multi-agent systems for enabling iot and
cps,” in Proceedings of the international conference on web intelligence.
places are working properly. All these properties ensure the ACM, 2017, pp. 356–364.
correctness of our system. We have also achieved the inter [12] C. Shum, W. H. Lau, T. Wong, T. Mao, S. Chung, C. Tse, K. F. Tsang,
operability and a well-defined process by using FIPA-ACL and L. L. Lai, “Modeling and simulating communications of multiagent
standards in our application. systems in smart grid,” in 2016 IEEE International Conference on Smart
Grid Communications (SmartGridComm). IEEE, 2016, pp. 405–410.
VI. C ONCLUSION [13] A. Pla, P. Gay, J. Meléndez, and B. López, “Petri net-based process
monitoring: a workflow management system for process modelling and
In this research, we have formally modeled FIPA perfor- monitoring,” Journal of Intelligent Manufacturing, vol. 25, no. 3, pp.
matives in RTMAS using TAPN for agent communication. 539–554, 2014.
Communication of agents is a significant characteristic of [14] S. Khosravifar, “Modeling multi agent communication activities with
RTMAS and is useful for message interaction in real-time petri nets,” International Journal of Information and Education Tech-
multi-agent systems. Previous work has focused on formal nology, vol. 3, no. 3, p. 310, 2013.
modeling of domain functionality of multi-agent systems and [15] B. Marzougui, K. Hassine, and K. Barkaoui, “A new formalism for
modeling a multi agent systems: Agent petri nets,” Journal of Software
not on the agent’s interaction level. The formal specification Engineering and Applications, vol. 3, no. 12, p. 1118, 2010.
and verification of these multi-agent systems in which the [16] J. F. Jensen, T. Nielsen, L. K. Oestergaard, and J. Srba, “Tapaal and
agents interact with one another to accomplish their objectives reachability analysis of p/t nets,” in Transactions on Petri Nets and
with time constraints ensures their reliability. TAPAAL has Other Models of Concurrency XI. Springer, 2016, pp. 307–318.
been used for the verification of the properties of interest of [17] M. Andersen, H. G. Larsen, J. Srba, M. G. Sørensen, and J. H.
the system specified through AF, AG, EG and EF fragments Taankvist, “Verification of liveness properties on closed timed-arc
of TCTL. The research provides future directions for formal petri nets,” in International Doctoral Workshop on Mathematical and
Engineering Methods in Computer Science. Springer, 2012, pp. 69–81.
modeling of standardized agent’s interaction with timing con-
[18] G. Guo, L. Ding, and Q.-L. Han, “A distributed event-triggered trans-
straints. The approach ensures that the system is deadlock free mission strategy for sampled-data consensus of multi-agent systems,”
and live. For the future, we will work on FIPA performatives Automatica, vol. 50, no. 5, pp. 1489–1496, 2014.
with Timed Colored Petri-nets in RTMAS. [19] G. S. Seyboth, D. V. Dimarogonas, and K. H. Johansson, “Event-based
broadcasting for multi-agent average consensus,” Automatica, vol. 49,
R EFERENCES no. 1, pp. 245–252, 2013.
[1] N. R. Jennings, K. Sycara, and M. Wooldridge, “A roadmap of agent re- [20] W. Louhichi, B. Marzougui, and K. Hassine, “Formal model for
search and development,” Autonomous agents and multi-agent systems, coordination in multi-agents system based petri net agent,” in 2017
vol. 1, no. 1, pp. 7–38, 1998. International Conference on Smart, Monitored and Controlled Cities
(SM2C). IEEE, 2017, pp. 134–137.
[2] B. Marzougui and K. Barkaoui, “Interaction protocols in multi-agent
systems based on agent petri nets model,” Int J Adv Comput Sci Appl, [21] P. G. Jensen, K. G. Larsen, and J. Srba, “Discrete and continuous
vol. 4, no. 7, 2013. strategies for timed-arc petri net games,” International Journal on
Software Tools for Technology Transfer, vol. 20, no. 5, pp. 529–546,
[3] A. Qasim, S. A. R. Kazmi, and I. Fakhir, “Formal specification and
2018.
verification of real-time multi-agent systems using timed-arc petri nets,”
Adv. Elect. Comput. Eng., vol. 15, no. 3, pp. 73–78, 2015. [22] J. A. Mateo, J. Srba, and M. G. Sørensen, “Soundness of timed-arc
workflow nets,” in International Conference on Applications and Theory
[4] A. Qasim and S. A. R. Kazmi, “Mape-k interfaces for formal modeling
of Petri Nets and Concurrency. Springer, 2014, pp. 51–70.
of real-time self-adaptive multi-agent systems,” IEEE Access, vol. 4,
pp. 4946–4958, 2016. [23] A. Kaur and S. Jain, “Kqml-from scenario to technology,” International
Journal of Advanced Studies in Computers, Science and Engineering,
[5] I. Obaid, S. A. R. Kazmi, and A. Qasim, “Modeling and verification
vol. 7, no. 3, pp. 30–34, 2018.
of payment system in e-banking,” INTERNATIONAL JOURNAL OF
ADVANCED COMPUTER SCIENCE AND APPLICATIONS, vol. 8, [24] W. M. Zuberek, “Timed petri nets definitions, properties, and applica-
no. 8, pp. 195–201, 2017. tions,” Microelectronics Reliability, vol. 31, no. 4, pp. 627–644, 1991.
[6] J. Pitt and A. Mamdani, “Some remarks on the semantics of fipa’s [25] J. Byg, K. Y. Jørgensen, and J. Srba, “Tapaal: Editor, simulator
agent communication language,” Autonomous Agents and Multi-Agent and verifier of timed-arc petri nets,” in International Symposium on
Systems, vol. 2, no. 4, pp. 333–356, 1999. Automated Technology for Verification and Analysis. Springer, 2009,
pp. 84–89.
[7] L. Chang, X. He, and S. M. Shatz, “A methodology for modeling multi-
agent systems using nested petri nets,” International Journal of Software [26] P. D. O’Brien and R. C. Nicol, “Fipa—towards a standard for software
Engineering and Knowledge Engineering, vol. 22, no. 07, pp. 891–925, agents,” BT Technology Journal, vol. 16, no. 3, pp. 51–59, 1998.
2012. [27] N. Dragoni and M. Gaspari, “Performative patterns for designing
[8] D. Juneja, A. Jagga, and A. Singh, “A review of fipa standardized agent verifiable acls,” in International Workshop on Cooperative Information
communication language and interaction protocols,” Journal of Network Agents. Springer, 2006, pp. 375–387.
Communications and Emerging Technologies, vol. 5, no. 2, pp. 179–
191, 2015.
www.ijacsa.thesai.org 537 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Jihad H’roura1 , Aissam Bekkari2 , Patrice Méniel5 Alamin Mansouri6 , Michaël Roy7
Driss Mammass3 , Ali Bouzit4 ARTEHIS laboratory, Le2i Laboratory
IRF-SIC Laboratory University of Bourgogne Franche-Comté University of Bourgogne Franche-Comté
Ibn Zohr University Dijon, France Auxerre, France
Agadir, Morocco
Abstract—Until the last decades, researchers taught that assume that data in human brain travels in the neural networks
teaching a computer how to recognize a bunny, for example, where each node performs a separate task, to achieve the visual
in a complex scene is almost impossible. Today, computer vision cortex where the recognition is performed based on its previous
system do it with a high score of accuracy. To bring the real world trained data. So, inspiring from this discovery, researchers in
to the computer vision system, real objects are represented as computer vision give another perspective called deep learning.
3D models (point clouds, meshes), which adds extra constraints
Concerning classical hypothesis, different approaches have
that should be processed to ensure a good recognition, for
example the resolution of the mesh. In this work, based on the been proposed depending on the explored levels of the object
state of the art method called Spin Image, we introduce our and also extracted features. If the targeted level is global and
contribution to recognize 3D objects. Our motivation is to ensure tends to describe the overall shape of the object, we talk
a good recognition under different conditions such as rotation, about global approaches. In the other hand, if the method
translation and mainly scaling, resolution changes, occlusions and focuses on extracting only local features, they are called local
clutters. To that end we have analyzed the spin image algorithm methods. Proposed approaches, either local or global aim to
to propose an extended version robust to scale and resolution ensure the robustness to different condition 3D objects in
changes, knowing that spin images fails to recognize 3D objects real scenes can be through for example, rotation, translation,
in that case. The key idea is to approach the representation of geometric deformations, occlusions, clutters, scaling, etc. In
spin images of the same object under different conditions by the
this respect, when it comes to occlusions and clutters, local
mean of normalization, either these conditions result in linear or
non-linear correlation between images. Our contribution, unlike approaches are known to be more efficient. Other strengths
spin image algorithm, allows to recognize objects with different of this category is that they are popular to not requiring any
resolutions and scale. Plus it shows a good robustness to occlusions segmentation and the pose estimation is simpler. However, the
up to 60% and clutters up to 50%, tested on two datasets: fact that local approaches are founded on local neighborhood,
Stanford and ArcheoZoo3D. which is highly affected by the resolution changes, make them
less discriminating. In addition, a verification, step is always
Keywords—3D object; recognition; spin image; resolution;
scaling needed to eliminate incorrect correspondences and the spatial
information is missed. Concerning global methods, they are
more discriminating since they provide a global description
I. I NTRODUCTION of the shape of the object. Besides, by only computing the
New information and communication technologies have nearest neighbor of the descriptor, we can perform matching,
emerged in 1990s and have grown exponentially in power. which makes it easier. In the opposite to local methods, they
The digital revolution which has been developing since its do not handle occlusions and clutters, the pose estimation is
birth at the end of 20th century, has affected different sectors more complicated and they usually require a segmentation as a
throughout the world. This revolution has led to the emergence pre-processing. In this paper we introduce a novel local shape
of new type of data, resulting in new and broader databases, based approach approach for 3D object recognition, crafted
particularly 3D data. This requires technological advances in to deal with resolution and scaling changes of the object
image processing or overall computer vision. Given the very in occluded and cluttered scenes. Our contribution, called
wide spectrum of industrial, military and medical applications Invariant to scale and resolution spin images (ISRSI), is based
that can be considered, this field in its turn has developed very on a state of the art method called spin images. Spin images
quickly. In the context of this digital revolution, notably in cog- fails when the resolution and the scale of objects change.
nitive sciences, scientists in computer vision have redirected By performing a normalization step and defining efficiently
their efforts to put into place a variety of interactive applica- the required parameters, we succeed to make this descriptor
tions with 3D real world, like 3D object recognition. To that invariant to scale and resolution changes. Our contribution has
aim, a better understanding of how the human visual system shown good robustness to occlusions up to 60% and clutters up
works is necessary. A first classical hypothesis assumes that, to 50%. The paper is laid out as follow. We briefly quote some
to recognize an object, the human brain starts by extracting related works in Section 2. Then, in Section 3 we describe the
features from objects captured by eyes. Then, depending on his background method. The Section 4 is dedicated to introduce
previous knowledge, he elaborates a matching process. Nev- more details about our contribution. While experiments are
ertheless, with the development of neuro-sciences, scientists conducted in Section 5. And finally, a conclusion is given in
www.ijacsa.thesai.org 538 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Section 6.
Fig. 4. Two spin maps and their two spin images of bunny with different
resolution.
Fig. 9. Two different scenarios of scaling of bunny: (a) The original object.
(b) Scaling and resolution scaling changes. (c) Scaling changes only.
So, the idea here is to bring the two spin images to the same
range [0,1], in order to normalize bin values for all spin images
to be able to compute the correlation coefficient efficiently.
In Fig. 8, we show the impact of normalization in the cor-
relation diagram of two spin images with different resolution
after normalization.
Here we can see clearly that the two images are more
correlated.
Fig. 7. The corresponding histograms of the two spin images from bunny
under different resolutions. In the histogram left, range of intensity values
varies between -5 and almost 20. Right, values are between -1 and almost 4. B. Invariance to Scale
One other drawback of spin image algorithm is the scaling.
We have two scenarios about scaling. The first one concerns
to show the difference between intensities of two spin images the object with the same number of vertices, but the scale is
that are meant to be similar. different, see Fig. 9(c) and the second one is when the same
object is represented with different resolution and scale in the
In Fig. 7, we illustrate their corresponding histograms to scene, see Fig. 9(b).
show clearly the difference between ranges.
The first case is simpler. As the scaling here does not
As the histogram is of essential importance in terms of change the normal vectors of vertices, the number of vertices
characterizing the global appearance of a given image one to project controlled by the parameter A (Angle support) is
needs to represent the values of compared histograms in the the same on each spin map. Since the image width is fixed for
same range in order perform an effective comparison. As both spin image model and scene, to deal with changes which
known, the min-max normalization approach is the simplest influences the accumulation of points in each bin of the spin
normalization technique in which we fit the data, in a pre- image, the bin size of the scene spin image should be set to
www.ijacsa.thesai.org 541 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 11. The impact of varying bin size on the distance match.
Fig. 10. Two spin images and their corresponding histograms of bunny and
its scaled version with no resolution changes. On left the spin image of
bunny and its histogram. On the right side the spin image of the scaled
version and its histogram.
bs = λbm (9)
Fig. 14. Objects used to run tests. three first objects from 3D Stanford
repository. Five second objects from ArcheoZoo3D database. And lastly
glove model.
Fig. 13. The influence of varying the bin size on the match correlation for domain, Precision and Recall. The mathematical formula for
bunny and caudal.
each one is given in equations (10) and (11).
tp
P recision = (10)
database. Afterwards, in Section B, we provide detailed tech- tp + f p
nical information on the implementation environment. Next,
the experiment carried out is revealed in Section C. In the tp
same section we measure the precision and recall to evaluate Recall = (11)
tp + f n
the performance of our contribution. And then we compare it
to the standard algorithm with a discussion of strengths and
shortcomings of our contribution. With tp: True positives is the number of times an existing
object in different scenes with different conditions is correctly
recognized. f p : False positives indicates the number of times
A. Datasets a non-existing object in the scenes is mentioned to be recog-
In this works we have validated our approach on two nized. It is to say that the algorithm finds correspondences on
datasets. The first one is Stanford 3D scanning repository. A the scene, so the model is aligned with another object. Finally,
well known repository that provides some dense polygonal f n: false negatives, when a model exists in the scene, but the
models publically. The second database is Archeozoo3D. It algorithm fails to recognize them, in our case, it fails to find
gathers 3D scans of horse’s bones. Before recognition we have any correspondences. To test the validity of our approach, we
processed objects to remove all unreferenced vertices. Then we used three objects from the Stanford repository, five objects
construct proper triangulated surfaces with screened Poisson from the ArcheoZoo3D database and one other object called
surface method to remove holes. We sampled all objects to glove modeled by Alexander Masliukivaky. The objects are
have the same resolution. listed in Fig. 14.
At first all objects have the same resolution. Resolution
B. Implementation here refers to the median of the lengths of the edges between
the vertices. The tests were done first for each isolated object.
In order to put the algorithm of spin images into action, we
We initially change only the resolution and keep the scale
have based our implementation on the information provided in
fixed, then apply the transformations (translation, rotations)
the thesis work [16]. We have implemented the whole phases
as well as truncating parts of the objects. We next carry out
of the algorithm from descriptor extraction to verification
tests in reverse. We fixed the resolution and changed the scale.
passing by the matching in Matlab. Concerning models in the
Lastly, we change both resolution and scale. In the second
two databases, they have been processed, whether for creating
time we test the robustness of our method to occlusions and
scenes, normalizing vectors or applying transformations, etc.
clutters. To do that, we have created 30 scenes from 4 objects
with the aide of Meshlab, blender and using the “Toolbox
of Stanford datasets, then we have changed the resolution of
Graph” of Peyre 1 in Matlab. About environmental informa-
scenes two times, which results in 90 trials for each model.
tion, our experiments were carried out on a computer with 2.50
Then we have repeated the same process for Archeozoo3D
GHz Intel i7 processor and 16GB of memory.
datastet. So roughly, concerning SRISI we get 360 trials for
Stanford and 360 for Archeozoo3D. For SI, as mentioned
C. Results and Discussion earlier, the algorithm does not find any correspondence. For
The purpose if this current section is to provide an evalu- the results presented in this work, image width is set to 64,
ation of our proposed method SRISI in comparison with the the resolutions of models is set to 0.3, the bin size is 0.15 and
original one SI. In order to provide a robust evaluation, the the angle support equal 180. To show the effect of occlusions
state of the art presents different metrics. We have chosen two and clutters on our method, we will compute the recognition
of the most important ones utilized in the information retrieval rate in terms of occlusions and clutters. To do this, for each
scene of the 30 scenes created from the Stanford database, we
1 http://www.mathworks.com/matlabcentral/fileexchange/ run the recognition test. This will allow us to deduce the true
5355-toolbox-graph positives, false positives and false negatives. Then we calculate
www.ijacsa.thesai.org 543 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
www.ijacsa.thesai.org 544 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
www.ijacsa.thesai.org 545 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Aisha Nazir1 , Muhammad Yaseen Khan2 , Tafseer Ahmed3 , Syed Imran Jami4 , Shaukat Wasi5
Center for Language Computing, Department of Computer Science, Mohammad Ali Jinnah University, Karachi, Pakistan1,2,3,4,5
Department of Computer Science, Edwardes College, Peshawar, Pakistan1
Abstract—Chatbots or conversational agents are the most to this question, we can say, the whole field of artificial
projecting and widely employed artificial assistants on online intelligence (AI) evolved. In this era, cognitive science deeply
social media. These bots converse with the humans in audio, observes the human’s mind and its working, which leads to
visual, or textual formats. It is quite intelligible that users are outstanding success in the field of AI in the form of artificial
keen interested in the swift and relatedly correct information for assistant aka Chabot. Businesses started to use these chatbots
their hunt in pursuit of desired product, such that their precious
time is not wasted through surfing multiple websites and business
to facilitate customers. Hence, the techniques and research
portals. In this paper, we present a novel incremental approach aspects of AI chatbots have become an exciting field in the AI
for building a chatbot for fashion brands based on the semantic community. These AI assistants/chatbots have revolutionized
web. We organized a dataset of 5,000 question and answers of top- by understanding customer queries in different languages and
10 brands in the fashion domain, which covers the information appropriately responding the meaningful information.
about new arrivals, sales, packages, discounts, exchange/return
policies, etc. We have also developed a dialogue interface for The main aim for these chatbots is to provides immediate,
querying the system. The results generated against the queries meaningful, informative, context-oriented responses to assist
are thoroughly evaluated on the criteria of time, context, history, customers for the asked questions. The AI Chatbots retrieve
duration, turns, significance, relevance, and fall back questions. information through different approaches. In modern-day prac-
Keywords—Artifical intelligence; semantic web; chatbots; fash- tices, these approaches use various information repository
ion; ontology structures like conventional (relational) and modern (NoSQL)
database systems, ontologies, AIML, etc. to model querying
I. I NTRODUCTION system.
The invention of the Internet has met almost everything In this paper, we present an ontology-driven chatbot model
in the world. It has played a significant part in showcasing that facilitates those customers who need the latest information
and growth of many businesses in many aspects [1], [2]. In about brands facilities such as packages, discounts, sales,
the context of the current era, online social media has made prices, varieties, online shopping, home delivery services,
a considerable impact on the businesses [3]. In the same etc. The proposed Chatbot covers all necessary and general
regard, the fashion industry, especially fashion brands that offer information relevant to clothing brands like dress designs,
voguish couture and apparels, is growing at a breathtaking fabric stuff, the material used in the product, accessories,
rate as they provide creative and versatile garments all around and services like home delivery, return, exchange, discounts,
the globe. People prioritize these brands upon their choices sales, and, etc. Through our model, customers will get all
and interests, which do have correlations with the events and type of information for their complex queries at one platform.
seasons. For example, customer can ask like: “Which brand provides
It is a challenging task to recommend an appropriate brand clutches in blue” and “What is the delivery time of Khaadi in
according to users’ requirements and interests. For doing so, Pakistan?”, etc. In this respect, we create an ontology-based on
there are many conversational agents available on the official the set of 5000 questions and answers considering the top-10
websites of clothing brands, but they deal with only one brand clothing brands of Pakistan, namely,
that resides it. What if a customer wants multiple options of the
same kind of different brands at one place? Yes, online social
• Asim Jofa • Thredz
media has answered this particular question to some extent,
but surfing the Internet to acquire desired results is very time- • HSY Studio • Gul Ahmed
consuming and exasperating effort. Thus, in comparison to the
searching and sorting based tools, people need some promising • Al-Karam Studio • Khaadi
alternatives [4]. • Sana Safina • J. (Junaid Jamshed)
To overcome user’s requirements to complete their task • Ethnic • Nishat Linen
with no time, it is mandatory to understand how human thinks
about a particular thing, in connection with this, it is also
imperative to understand that how do we make computers
to do it like humans. Turing first floated this idea, “Can We are hopeful that the proposed model is adept for many
a machine think?” [5], [6], and in pursuit of the answer other but similar domains, all over the world.
www.ijacsa.thesai.org 546 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
owl:Thing
Brand/Vendor
Accessories
FemaleAccessories
MaleAccessories
Fig. 1. The scheme of ontology engineering process (OEP). Block enclosed BrandType
in green box is the initial phase of data gathering, while the block in red ClothVariety
box defines the OEP.
FemaleClothVariety
MaleClothVariety
DressCategory
users and different teams. Hence, we have employed a Google Facility
Chrome extension, namely, Scrapper1 , for scraping the in- FemaleBrands
formation from the web-portals. The other tool used, in our LatestDesignVolume
course of the experiment, for extracting the text of posts and Location/Area
comments from Facebook pages, namely, FacebookPager2 . The MaleBrands
FacePager takes Facebook page key as an input and retrieves Packages
all posts, comments, pictures, videos, and user reviews avail- PriceRange
able on the given page that you can export in comma-separated Scope
view format. Timing
Fig. 2. Class Hierarchy of Fashion Brand Ontology
As a result, the collected corpus was consisting of un-
structured and inconsistent data. Further, there were two more
issues: it was redundant and not very much meaningful. Thus,
the data is filtered and processed to make it useful as per the 4) Properties, Attributes, or Predicates: Since these set of
requirements of the SW based applications. The count of the classes are not self-explanatory, therefore, we have to define
parallel corpus, in two-way communication, is 5,000 sentences. the mappings inside/among classes [29]. The OWL properties
describe the relationship between classes which can be of two
2) Competency Questions: After the acquisition of data, the kinds, namely, object properties, and data properties. Details
first step towards ontology development is to lemmatize the of these properties are given below.
scope of ontology through the competency questions. These
are the vital questions for which an ontology has to answer. Object Properties. These properties are ones who estab-
Moreover, these questions are the primary source of setting lish a link between two individuals. This is also known as
the precincts of ontology domain, and helpful to identify the intrinsic properties [7]. Technically, as a rule of thumb, any
terms that are further converted into the system of class and property whose range is a class is an object property. Protégé
subclass hierarchy. In the proposed domain, for example, the provides numerous predicates that remove ambiguity from the
competency questions can be: “Does Asim Jofa provide ex- taxonomy. Fig. 3 shows the object properties of the fashion
change/return facility?”, “Which brand provides accessories?”, brand ontology.
or “Which brand offers 50% discount?”, etc.
owl:TopObjectProperty
3) Concepts and Classes: Classes are the basic building
IntroducePackage
block of ontology, which can be interpreted as a set of specific
Labelled_as
individuals [28]. In OWL, these are also called concepts or
Located
entities having some distinctive characteristics. These classes
OfferAccessories
are formed in a hierarchical system of super-class and sub-
PresentDressCategory
class. However, these classes can be disjoint. In such case,
ProposeClothVariety
the individuals of these classes are not common. This class
ProvideFacilities
and sub-class hierarchy is also known as Taxonomy [7]. Thus
Releases
every super class exhibits the most general characteristics of
Type
all nested sub-classes, and in contrast, the sub-classes does
opposite. For example, in our experiment, “brand/vendor” Fig. 3. Object Properties.
is the most general class which has nested sub-classes like
“accessories”, “cloth variety”, “brand type”, “dress category”,
etc. Likewise, the class “accessories” is further nested by two
The core characteristics of object properties which show
1 https://chrome.google.com/webstore/detail/scraper/ the global cardinality constraints on properties are: Func-
mbigbapnjcgaffohmbkdlecaccepngjd tional, Inverse functional, Transitive, Symmetric, Asymmetric,
2 https://github.com/strohne/Facepager/wiki/About-Facepager Reflexive, and Irreflexive properties. Intuitively, functional
www.ijacsa.thesai.org 548 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
property is a property which postulates that for any given Data Properties. We can briefly define a data property as
individual there must be at most one out going relationship a property that relates individuals to data-type values [33], in
[22], [28], [30]; inverse property asserts that for any individual other words, any property whose range is any literal or data-
there should be at most one incoming relationship, through type value is known as a data property. Extrinsic properties:
the property, which can uniquely identify the subject [30], see like name, has string data-type. Table II shows the details of
Fig. 4b, where a brand/vandor offers accessorises, which can data properties with domain and ranges accordingly.
identify the provider inversely. However, if there are many
things related to one individual, through the functional or
TABLE II. E XAMPLES OF DATA PROPERTIES
inverse-functional property, then the property characteristic
will be inconsistent [22], [28]. The transitive property can be Property Name Domain Range
defined as the property which shows transitive implications Is a Levise maleBrand
Is a Bareeze femaleBrand
among individuals, such that if an individual a is similar to BrandTypeIs Nishat Luxury
individual b, and b is similar to individual c than we can say Functional
HasStuffQuality Khaddi Moderate
AccessoryProvidedBy Scarfs JunaidJamshaid
the individual a and c are also similar, through a transitive AccessoryProvidedBy Belt Levise
relation [30]–[33]. Fig. 4c depicts an example w.r.t to the OfferDressCategor SanaSafina Bridal
current research. The symmetric property asserts that a given LabledAs Khaddi International
HasScope Adidas International
individual has itself an inverse function; whereas asymmetric IsTypeOf Bareeze Luxury
property lacks this characteristic [30], [32]. We can see, as an Asymmetric
LocatedAt Khaddi Saddar
example of symmetric property, if individual a and individual b OfferedDiscount Bonanza float
OfferAccessories Brand/Vendor Accessories
are related to each via some property, then b should be related ProvidedBy Accessories Brand/Vendor
to a through the same property; while in the same setting, for LabledAs Brand/Vendor Scope
the asymmetric case, b does not relate to a along the same IsScopeOf Scope Brand/Vendor
Inverse
PresesntDressCategory Brand/Vendor DressCategory
property. Fig. 4d and 4e show the examples of symmetric DressCategoryOfferedBy DressCategory Brand/Vendor
and asymmetric properties respectively. Lastly, the reflexive ProposeClothVariety Brand/Vendor ClothVariety
ClothVarietyOfferedBy ClothVariety Brand/Vendor
property relates everything to itself, whereas, the irreflexive
property means no individual can be related to itself by some
role [22], [32]. Few examples of these object properties are 5) Instances: An instance is an individual/object that cer-
illustrated in Fig. 4, and detailed mapping of these properties tainly belongs to a class. One key feature of OWL ontology is:
is shown in Table I. it does not use the unique name assumption (UNA), so we can
explicitly define that two individuals are the same or different.
A class may have multiple instances. We can manually define
the characteristics of each instance separately. For example,
TABLE I. E XAMPLES OF O BJECT P ROPERTIES
class Brand has instances like Khaddi, AsimJofa, Nishat, Al-
Property Name Example Karam, and many others.
(maleBrand, femaleBrand)
Symmetric IsSibblingOf (maleAccessories, femaleAccessories 6) Axioms: After building class taxonomy and establishing
(maleClothingVariety, femaleClothingVariety)
SamaFacilitiesAs (Khaddi,Nishat,GulAhmed)
links among classes and individuals: the following step is
SamaClothVarietyAs (Khaddi,Nishat,GulAhmed) carried out to the semantics unambiguous. It is done so to
SamePriceRangeAs (Nishat,JunaidJamshed,GulAhmed) ensure the validation and consistency of ontology; and as a
Transitive SamePacksgesAs (Thredz,Levise,Nishat)
SameAccessoryAs (Bonanza,GulAhmed,Nishat) procedure, we convert hierarchy into first-order logic, hence
SameClothingStuff (Nishat,Khaddi,Bonanza) forming the “axioms” that is represented as hC, R, I, Ai, where
SameDressCategoryAs (Bonanaza,GulAhmed,Khaadi) C represents classes, R represents relations therein, I shows
www.ijacsa.thesai.org 549 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
their instances, and A shows axiom [26]. Protégé provides to produce context-oriented responses. In this regard, IR
‘Reasoner’ support to formulate and manipulate logical formu- approaches are commonly practiced, in connection with the
lae [33]. Through Reasoner, we can check the consistency of combination of rules. Thus, based on rules, a history-oriented
ontology; as well as add inference to semantic web application and well-aligned conversation leads to generate more accurate
[8]. Further, in order to visualize ontology with its conceptu- and logical responses. Basic programming structures like con-
alization, graphical representation was generated using Onto- ditional statements and repeating structures are employed.
Graf and VOWL plugins [29]. Fig. 5 shows the visualization
of the proposed ontology. C. Integration and Querying with SPARQL
On completing this phase, we are done with the ontology In this phase, we deploy ontology in an environment
engineering process; thus, in next sections, ontology integra- where it can easily retrieve data by establishing a connection
tion and chatbot designing phases are discussed in detail. between the chatbot interface and itself. “Jena”3 is a Java-
based library, specifically use to support semantic web-based
B. Rule Engineering applications [26]. Semantic query language (SPARQL) is used
to manipulate semantic web repository [25]. SPARQL operate
We worked out different scenarios based rules at the back- on the triple store, and itself entails triple pattern. Protégé
end of the chatbot. These rules are defined to make the
conversation in a flow, and the system to be more efficient 3 https://jena.apache.org/
www.ijacsa.thesai.org 550 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
core facility provided by international brands. Several tussles [16] G. S. Pathan, P. M. Bante, R. V. Dhole, and S. Kurzadkar, “An e-
are required to make a well organized artificial bot to produce commerce web application based chatbot,” in International Journal
fast results. The proposed system resolves the problem for for Research in Applied Science & Engineering Technology, 2018, pp.
1263–1267.
Pakistani fashion industry through developing clothing brand
[17] A. Augello, G. Pilato, A. Machi, and S. Gaglio, “An approach to
ontology, yielded through the handcrafted dataset of 5000 pairs enhance chatbot semantic power and maintainability: experiences within
of questions/answers, and integrating it with a conversation the frasi project,” in 2012 IEEE Sixth International Conference on
agent to facilitate online customers. In our work, we focus only Semantic Computing. IEEE, 2012, pp. 186–193.
on general-purpose information like brand facilities, services, [18] Y. Wu, W. Wu, C. Xing, M. Zhou, and Z. Li, “Sequential matching net-
garments, clothing stuff, and accessories based on information work: A new architecture for multi-turn response selection in retrieval-
based chatbots,” arXiv preprint arXiv:1612.01627, 2016.
retrieved from Facebook pages and official websites.
[19] J. H. Gennari, M. A. Musen, R. W. Fergerson, W. E. Grosso,
M. Crubézy, H. Eriksson, N. F. Noy, and S. W. Tu, “The evolution of
VI. L IMITATIONS AND F UTURE W ORK protégé: an environment for knowledge-based systems development,”
International Journal of Human-computer studies, vol. 58, no. 1, pp.
This research work is limited to only ten clothing brands 89–123, 2003.
and provides concern areas information to customers; thus, in [20] G. Klyne and J. J. Carroll, “Resource description framework (rdf):
the future, the scope of brand ontology can be increased by Concepts and abstract syntax,” 2006.
adding more national brands. We also intend to implement [21] G. Antoniou and F. Van Harmelen, “Web ontology language: Owl,” in
Semantic Web Rule Language (SWRL) and employing deep Handbook on ontologies. Springer, 2004, pp. 67–92.
learning architectures. [22] D. L. McGuinness, F. Van Harmelen et al., “Owl web ontology language
overview,” W3C recommendation, vol. 10, no. 10, p. 2004, 2004.
R EFERENCES [23] S. Lohmann, S. Negru, F. Haag, and T. Ertl, “Visualizing ontologies
with vowl,” Semantic Web, vol. 7, no. 4, pp. 399–419, 2016.
[1] B. Gates, “Business@ the speed of thought,” Business Strategy Review, [24] S. Falconer, “Ontograf,” Protégé Wiki, 2010.
vol. 10, no. 2, pp. 11–18, 1999.
[25] T. Segaran, C. Evans, and J. Taylor, Programming the Semantic Web:
[2] G. J. Avlonitis and D. A. Karayanni, “The impact of internet use on Build Flexible Applications with Graph Data. ” O’Reilly Media, Inc.”,
business-to-business marketing: examples from american and european 2009.
companies,” Industrial Marketing Management, vol. 29, no. 5, pp. 441–
[26] B. McBride, “Jena: A semantic web toolkit,” IEEE Internet computing,
459, 2000.
vol. 6, no. 6, pp. 55–59, 2002.
[3] S. Aral, C. Dellarocas, and D. Godes, “Introduction to the special
[27] J. J. Carroll, I. Dickinson, C. Dollin, D. Reynolds, A. Seaborne,
issue—social media and business transformation: a framework for
and K. Wilkinson, “Jena: implementing the semantic web recommen-
research,” Information Systems Research, vol. 24, no. 1, pp. 3–13, 2013.
dations,” in Proceedings of the 13th international World Wide Web
[4] S. Gupta, D. Borkar, C. De Mello, and S. Patil, “An e-commerce conference on Alternate track papers & posters. ACM, 2004, pp.
website based chatbot,” International Journal of Computer Science and 74–83.
Information Technologies, vol. 6, no. 2, pp. 1483–1485, 2015.
[28] M. Horridge, S. Jupp, G. Moulton, A. Rector, R. Stevens, and C. Wroe,
[5] A. Turing, “Can machine think?” “A practical guide to building owl ontologies using protégé 4 and co-ode
[6] L. S. C. Rao, D. Kini, K. S, and K. K. N, “Chatbot-a java based tools edition1. 2,” The university of Manchester, vol. 107, 2009.
intelligent conversational agent,” International Research Journal of [29] K. Hadjar, “University ontology: A case study at ahlia university,” in
Engineering and Technology, vol. 4, no. 4, pp. 3575–3578, 2017. Semantic Web. Springer, 2016, pp. 173–183.
[7] N. F. Noy, D. L. McGuinness et al., “Ontology development 101: A [30] T. B. of Trustees of the Leland Stanford Junior University. Protégé
guide to creating your first ontology,” 2001. 5 documentation. [Online]. Available: http://protegeproject.github.io/
[8] S. A. Abdul-Kader and J. Woods, “Survey on chatbot design techniques protege/
in speech conversation systems,” International Journal of Advanced [31] P. Hitzler, M. Krötzsch, B. Parsia, P. F. Patel-Schneider, and S. Rudolph,
Computer Science and Applications, vol. 6, no. 7, 2015. “Owl 2 web ontology language primer,” W3C recommendation, vol. 27,
[9] T. Berners-Lee, J. Hendler, O. Lassila et al., “The semantic web,” no. 1, p. 123, 2009.
Scientific american, vol. 284, no. 5, pp. 28–37, 2001. [32] W. W. W. Consortium et al., “Owl 2 web ontology language document
[10] C. Bizer, “The emerging web of linked data,” IEEE intelligent systems, overview,” 2012.
vol. 24, no. 5, pp. 87–92, 2009. [33] A. Grigoris and H. F. Van, “A semantic web primer a semantic web
[11] A. Bernstein, J. Hendler, and N. Noy, “A new look of the semantic primer second edition,” 2008.
web,” 2016. [34] K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “Bleu: a method
[12] K. Bollacker, N. Dı́az-Rodrı́guez, and X. Li, “Beyond clothing ontolo- for automatic evaluation of machine translation,” in Proceedings of
gies: modeling fashion with subjective influence networks,” in KDD the 40th annual meeting on association for computational linguistics.
workshop on machine learning meets fashion, 2016. Association for Computational Linguistics, 2002, pp. 311–318.
[13] D. Vogiatzis, D. Pierrakos, G. Paliouras, S. Jenkyn-Jones, and B. Possen, [35] S. Banerjee and A. Lavie, “Meteor: An automatic metric for mt evalua-
“Expert and community based style advice,” Expert Systems with tion with improved correlation with human judgments,” in Proceedings
Applications, vol. 39, no. 12, pp. 10 647–10 655, 2012. of the acl workshop on intrinsic and extrinsic evaluation measures for
[14] S. Ajmani, H. Ghosh, A. Mallik, and S. Chaudhury, “An ontology machine translation and/or summarization, 2005, pp. 65–72.
based personalized garment recommendation system,” in Proceedings [36] C.-Y. Lin, “Rouge: A package for automatic evaluation of summaries,”
of the 2013 IEEE/WIC/ACM International Joint Conferences on Web in Text summarization branches out, 2004, pp. 74–81.
Intelligence (WI) and Intelligent Agent Technologies (IAT)-Volume 03. [37] H. Chen, X. Liu, D. Yin, and J. Tang, “A survey on dialogue systems:
IEEE Computer Society, 2013, pp. 17–20. Recent advances and new frontiers,” Acm Sigkdd Explorations Newslet-
[15] H. Al-Zubaide and A. A. Issa, “Ontbot: Ontology based chatbot,” in ter, vol. 19, no. 2, pp. 25–35, 2017.
International Symposium on Innovations in Information and Communi- [38] J. H. Martin and D. Jurafsky, Speech and language processing: An
cations Technology. IEEE, 2011, pp. 7–12. introduction to natural language processing, computational linguistics,
and speech recognition. Pearson/Prentice Hall Upper Saddle River,
2009.
www.ijacsa.thesai.org 552 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Nghia Duong-Trung1 , Xuan Son Ha2 , Tan Tai Phan3 , Phuong Nam Trieu4 ,
Quoc Nghiep Nguyen5 , Duy Pham6 , Thai Tam Huynh7 , Hai Trieu Le8
Can Tho University of Technology Can Tho city, Vietnam1,2,4,5,8
FPT University, Can Tho city, Vietnam1,2
National Chiao Tung University, Taiwan3
Hanoi University Science Technology, Ha Noi, Vietnam6
Transaction Technologies Pte. Ltd., Singapore7
Abstract—To date, cash on delivery (COD) is one of the COD has become increasingly popular in recent years and been
most popular payment methods in developing countries thanks considered one of the main payment methods in many coun-
to the blossom of customer-to-customer e-commerce. With the tries [1], [2], [3]. Among research articles, most investigated
widespread of a very small business model and the Internet, payment methods is in general, rather than focusing on COD
online shopping has become part of people’s daily activity. People in particular. Transfer agents are often used as postal services,
browse for desirable products at the comfort of their homes and
ask the online vendor that a shipper can deliver the merchandise
but usually, consumer and business shipments will be sent
at their doorstep. Then, COD allows customers to pay in cash to COD by courier companies, commercial truck forwarders
when the product is delivered to their desired location. Since or organizations own delivery services. COD sales usually
customers receive goods before making a payment, COD is, involve a delivered fee charged by the shipping agents and is
therefore, considered as a payment system. However, the crucial usually paid by the buyer. In retail and wholesale transactions,
issue that previous research has not yet addressed is that their shipments rely on COD-based payment method when the buyer
models only support single delivering session at a time. More does not have a credit account and the seller does not choose
precisely, if the current buyer is not available to receive the goods, a payment method in advance. COD postal services [4] were
the shipper has to wastefully wait for the complete payment and first introduced in Switzerland in 1849, India and Australia
he/she cannot start shipping another merchandise. The tracking in 1877, the United States in 1913, Canada in 1922 and the
system seems to poorly handle this issue. In particular, we
propose a multi-session mechanism, which consists of blockchain
United Kingdom in 1926. Particularly in Vietnam, COD is
technology, smart contracts and hyperledger fabric platform to accepted by almost online vendors and customers.
achieve distributed and transparent across delivering sessions
in the decentralized markets. Our proposed mechanism ensure
the efficiency of delivering process. The authors release our However, the crucial issue that previous research has not
sources codes for further reproducibility and development. We yet addressed is that their models only support single deliver-
conclude that the integration of multi-session mechanism and ing session at a time. During a working day, a shipper can take
blockchain technology will cause significant efficiency across as many orders from customers across the local area. Then,
several disciplines.
the shipper delivery the products sequentially. If the current
Keywords—Blockchain; cash on delivery; multi-sessions; decen- buyer is not available to receive the goods, the shipper has to
tralized system wastefully wait for the complete payment and he/she cannot
start shipping another merchandise. The tracking system seems
I. I NTRODUCTION to ineffectively track this issue. Consequently, a mechanism
is missing in the buyer and seller’s dilemma [5]. Addition
With the adoption of modern technology and the Internet, to the current seven core components, e.g. product delivery,
selling products online has become a very active market in product payment, delivery trust, payment trust, escrow account,
developing countries. There is an immense need to have a legal document and reputation system, the authors introduce a
delivery solution of any physical items ranging from catering, new part that can be integrated into the dilemma. To the best
beverages, clothing and home facilities. Meeting the needs of of our knowledge, this novel idea is firstly investigated and
today’s customer-to-customer e-commerce, many third parties implemented by the authors.
have launched delivery services which utilize blockchain tech-
nology. It seems counter-intuitive that, in developing countries
like Vietnam, credit card and online payment are not widely
The rest of the paper is organized as follows. The authors
used in the market. People still prefer to pay in cash because
present related research in the field of COD and decentral-
they want to make sure that the products must be in perfect
ized system in Section II. Then in Section III, the authors
condition.
summarize the most important technical background for com-
Cash on Delivery (COD) allows customers to pay in cash prehending the proposed mechanism. The core contribution of
when the products are delivered to their home or a location the paper is presented in Section IV. Next, several real-world
they choose. This is sometimes called a payment system scenarios and remarks are demonstrated in Section V. Finally,
because customers receive goods before making a payment. the authors make conclusion in Section VII.
www.ijacsa.thesai.org 553 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
II. R ELATED W ORK article has deployed a survey regarding influencing factors to
payment transactions. That surveys emphasize two factors, e.g.
Hanan and Salah have mentioned some limitations of
trust and security. Similar to the above survey, the payment and
proof of delivery (POD) process which relies on a trusted
transportation in COD need the trust of the participants and
third party to implement the process [6]. Therefore, a new
the security of the system.
POD process using Blockchain technology from Ethereum has
been introduced with the number of transportations through OpenBazaar platform [14] provides a procedure for making
several intermediaries by their research. Besides, a dual deposit deposits when agreed by the buyer, seller, and trader which is
mortgage mechanism is used for parties to comply with the known as multi-signature escrow. A third party will participate
contract. The development of current e-commerce and the in the process of trading an item called a moderator. The mod-
important role of Blockchain technology has been dedicated in erator will resolve disputes when a problem occurs. Bitcoin
Ha et al. research [7]. Besides, the limitation of the traditional currency will be used for payment in transactions. The process
CoD model is mentioned in detail such as trusted in the third has not yet delivered the person, the role of the deliverer is
party, order management, and the payment process between not specified. Besides, the need for a third party to resolve
the parties in the system. So that, Blockchain technology using the dispute will consume more assets and time of the parties
Hyperleder and smart contract is built to solve the issues of involved. This has been solved by using the smart contract as
COD. a third party as well as solving the problems of the parties
involved in the contract terms.
Camp et al. [8] had provided a digital purchasing method
with a digital token on the network. They offer to issue invoices COD model using two smart contracts is introduced by
signed by sellers and digital goods which have been encrypted Le et al. [15]. An outcome is a positive deployment process
and sent from a seller to a customer. The participants such of decentralized applications, which enforces contracts with
as sellers and buyers will be anonymous, making transactions the exact terms. The price of the order is deposited by the
through commitments by signing confirmation. There are no participants. However, the management of orders becomes
legal or property constraints. Le et al. [9] has mentioned the more difficult when the data and the number of orders scale
important roles of blockchain, especially, the decentralized up. According to the process, the second contract will be
users model, to builds the transportation process and offers implemented immediately after the first contract is executed.
mechanisms to promote and ensure the interests of partici- The implementation is based on the memorization of the
pating parties. The benefits of the seller are enhanced and address of each contract. As a result, it is a major limitation
penalized the shippers who deliberately cheated. Therefore, in the deployment process if the process is applied to multiple
the real data has applied to the system so that the delivery orders since the system could not perform several contracts at
of multiple senders suitable for their Blockchain system is the same time.
transparent. The process is built by all cash payments.
Altawy et al. [10] have compared the differences between III. M ATERIALS AND T ECHNICAL BACKGROUND
buying handicrafts using cash payment and buying goods A. Cash On Delivery
via the Internet. Online shipping needs more trust and the
information of the parties who join the system to perform COD is a service of collecting money collected in the
several actions such as making payment, delivery, and making group of services of buying goods by post (Cash On Delivery
sure the right items. Besides, the types of e-commerce used in or Collect On Delivery). It is the association between postal
trading which help the process faster. Anonymously purchase service and money transfer service with several stages: First,
of a buyer is a big concern so that the Lelantos system has the shipper proceeds to send the goods to the recipient via
built based on Blockchain to detect and cancel the anonymous a courier company. Next, the courier company will send the
purchases which affect on the trading process. Discussing goods to the recipient by delivery service. The branch or
trading on a digital platform with the trust of the participants, post office of the delivery company delivers the goods to
Asgaonkar and Krishnamachari [11] have issued a deposit the recipient and the receiver makes payment. After that, the
protocol for trading by the participants. They applied the trust branch or delivery office issued a COD check (similar to a
of the Blockchain system to make payment by participants money order) sent to the shipper. From this point on, COD
without a trusted third party. This protocol asked dual-deposit will be similar to a money transfer service.
amount on the contract with payment of both sides and the
price of the product is always fixed. However, the product for B. Blockchain Technology
the parties conducting the transaction has not been verified.
Blockchain is a list of continuously written logs, called
Halaweh presented the rapid growth of the COD model blocks, linked by encryption. Each block contains the previous
as an important method in making payment and transport in block’s cryptographic hash function, timestamp, and transac-
e-commerce [12]. The author gave statistics on the study of tion data. Each block has a block header and a body containing
the COD process to customers and conducted the prediction data and hash values of the previous block. The hash value is
and testing of factors affecting the COD process by using the result of a hash function. The hash function transforms
a questionnaire methodology. Moreover, it also predicts the data of any length into a fixed-length string or numeric value,
factors that affect the COD process such as safety and security such as 256 bits (32 bytes) with SHA256. Blockchain is a
of the system for the products and privacy of the participants. technology that allows secure data transmission based on an
Barkhordari et al. have proposed a concern of the bank using extremely complex encryption system, similar to accounting
the Internet to negotiate and solve customers’ needs [13]. That books of a company where cash is closely monitored. In
is the trust of customers and the security of the system. Their this case, the blockchain is an accounting ledger [16] that
www.ijacsa.thesai.org 554 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
works in the digital field. A special feature of blockchain the product information is uploaded to the sale contract by
is that transactions are done at a high level of trust without the seller where the buyer can verify through the app, and
disclosing information. All types of business and management send the purchase request. The sale contract will trigger the
can participate in the network and use the properties of the purchase contract for activating the term that the buyer has
Blockchain system to ensure transparency of stakeholders. to transfer the amount of mortgage money as same as the
valuation of the order to the purchase contract. Thus, a delivery
C. Ethereum request is sent to the system by the shipper after seeing
an available order. The sale contract will trigger to delivery
Ethereum [17], [16], [18] is a distributed, public, and open- contract for activating the terms that the shipper and seller
source computing platform based on blockchain technology. It have to transfer mortgage money and delivery fee, respectively.
features smart contracts (scenarios), facilitating online contract Finally, the money will be transferred to the seller from the
agreements. This platform includes a complete Turing virtual purchase contract. The mortgage money and delivery fee will
machine and Ethereum Virtual Machine [19], [20], which be transferred to the shipper by the delivery contract which
can execute scripts using an Ethereum computer network. used to store the delivery fee and order money from the seller
Ethereum also provides a cryptocurrency called Ether, which and shipper respectively.
can be transferred between accounts and used to pay peaches to
help perform calculations. Gas is an internal transaction pricing
mechanism, used to minimize spam and allocate resources on
the network. When creating, each transaction is charged by
a certain amount of gas, its purpose is to limit the amount
of work needed to execute the transaction and pay for this
execution at the same time.
D. Smart Contracts
A cryptocurrency is a decentralized platform that a dis-
tributed ledger is used to interact with virtual money. A
contract is an instance of a computer program that executes
on the Blockchain. Users transfer money by publishing trans-
actions and interacting with contracts in the cryptocurrency
network where information is propagated, data is stored among
miners or network’s nodes. An underlying cryptocurrency
system supports the utilization of smart contracts. A smart
contract contains program code, a stored file and an account
balance. Any user can submit a transaction to an append- Fig. 1. General description of our proposed multi sections of COD transport
able-only log. When the contracted is created, its program process.
code cannot be changed. An append-able-only log, called a
blockchain, which imposes a partial or total arrangement on
submitted transactions is the main interface provided by the
cryptocurrency. The integration of smart contract in COD has B. Detailed COD Scenarios
been discussed in [15]. 1) Shipper successfully delivers goods and the buyer suc-
cessfully receives the goods: The sale contract triggers to
E. Decentralized Applications – dApps purchase contract, seller contract and delivery contract for
activating the money transferability function. The order money
DApps [21], [22], [23] are as similar as normal applications will be transferred to the seller by purchase contract when the
except that they are completely decentralized. It is also con- buyer confirms successful delivery to the delivery contract.
trolled by nodes running Ethereum networks. These dApps do On the other hand, the delivery contract returns the mortgage
not depend on any central server or third party for operating, money to the shipper which is already deposited before the
and therefore, without the central point of failure. Thanks receiving the order, and the delivery fee is also transferred to
to the blockchain technology, the database is encrypted and the shipper by seller contract immediately. This scenario is
stored in a decentralized fashion. By using a modern mean of illustrated in Fig. 2.
communication protocols, participants can store and retrieve
data without the risk of censor and intervention [24]. DApps 2) Shipper unsuccessfully delivers goods: The sale contract
are expected to resist attack and censorship while being able triggers to purchase contract, seller contract and delivery
to operate in a fully autonomous model. contract for activating the money transferability function. The
order mortgage money and delivery money will be transferred
IV. P ROPOSED M ULTI -S ESSIONS COD P ROCESS to the seller by seller contract and delivery contract due to the
shipper failed delivery, then the purchase contract returns the
A. Abstract Model for a COD system mortgage money to the buyer. This scenario is illustrated in
Fig. 3.
The authors start this session by presenting a general
description of multi sections in COD transport process. The 3) Buyer refuses to receive goods: The sale contract trig-
abstract model for COD system is illustrated in Fig. 1. First, gers to purchase contract, seller contract and delivery contract
www.ijacsa.thesai.org 555 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Fig. 2. Case 1: Shipper successfully delivers goods and buyer successfully Fig. 4. Case 3: Buyer refuses to receive goods.
receives the goods.
The algorithm (2) is the refund algorithm. Line 1, 2, and 3 1: Trigger transfer money function
create temporary addresses. Triggering the seller contract to get
a mortgage is executed at line 4. Trigger purchase contract to
get back mortgage is done at line 5. Line 6 is to trigger delivery
contract to transfer the mortgage of the shipper’s products to 3) Case 3: The shipment is done successfully. The buyer
the seller. transfers money to the seller. The seller transfers money to the
shipper. The shipper takes the money. Line 1 in the algorithm
Algorithm 2 Refund algorithm (6) triggers the transfer money method.
Input: Order code
Output: Trigger the contracts to execute money transferability Algorithm 6 Case 3: The shipment is done successfully
and get back mortgage money Input: Order code
Output: Trigger transfer money method
1: Address seller deposit temp
2: Address buyer deposit temp 1. Trigger transfer money function
3: Address shipper deposit temp
4: Trigger DepositSeller(seller deposit temp) contract and
function to get back mortgage money 4) Case 4: The order is wrong because of the seller.
5. Trigger Depositbuyer(buyer deposit temp) contract and Shipper checks the order at line 1 and 2 in the Algorithm (7).
function to get back mortgage money When the order is wrong, the ReUnfundSellerFail function is
6. Trigger DepositShiper(shipper deposit temp) contract activated at line 3. The shipment stops unsuccessfully.
and shipper function to transfer money
Algorithm 7 Case 4: Seller is failed
The algorithm (3) is called the seller failure algorithm Line Input: Order code
1, 2, and 3 create the temporary addresses. Shipper checks the Output: Trigger the reunfundSellerFail method
order and returns the money if the order is not correct is done
at line 4. Delivery mortgage money of the seller is a return to 1: Trigger package [ order code ] name = name
the seller is executed at line 5. 2: If name != name
3: Trigger ReUnfundSellerFail function
1) Case 1: The failure is caused by the shipper. The 4: EndIf
mortgage money as same as the valuation of the order is
triggered to refund payment method as set at line 1 in the
algorithm (4).
V. E XPERIMENTS
2) Case 2: The failure is caused by the buyer. When the
situation happens, the shipper will receive the package and On a blockchain Ethereum model, all of the interaction
delivery fee. It is done at line 1 in the algorithm (5) to trigger with the blockchain such as contract reaction, command
the transfer money method. translation, execution of function has to pay a fee which
www.ijacsa.thesai.org 557 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
called gas. Gas costs depend on the complexity and logic of TABLE V. C ASE 1: S TEP 5 A
that function. It is calculated based on how much computer
resources will be required to perform the function. So that, From 0x4b0897b0513fdc7c541b6d9d7e929c4e5364d2db
code optimization is important in Ethereum to be able to save To DepositShipper.(constructor)
costs. The measurement in four experimental cases of COD has Transaction cost 239169
Execution cost 142813
also performed. Case 1, the process takes place normally, the
receiving and shipping take place successfully. Case 2, we will
refer to the transaction error due to the problem on the shipper. TABLE VI. C ASE 1: S TEP 5 B
Case 3 is a transaction error due to a buyer problem. Finally,
the seller delivers the wrong product, e.g. Case 4. The details From 0x4b0897b0513fdc7c541b6d9d7e929c4e5364d2db
implementation of these cases is presented in this section. A Seller.setShipperDepositAddress(uint256,address)
To
complete codes solution is publicized on the authors’ GitHub 0x692a70d2e424a56d2c6c27aa97d1a86395877b3a
Transaction cost 43443
repository1 (CC BY 4.0) to engage further reproducibility and Execution cost 20635
improvement.
From 0x4b0897b0513fdc7c541b6d9d7e929c4e5364d2db
TABLE I. C ASE 1: S TEP 1 Seller.setSellerDepositAddress(uint256,address)
To
0x692a70d2e424a56d2c6c27aa97d1a86395877b3a
From 0xca35b7d915458ef540ade6068dfe2f44e8fa733c
Transaction cost 43465
Seller.setPackage(string,uint256,string)
To Execution cost 20657
0x692a70d2e424a56d2c6c27aa97d1a86395877b3a
Transaction cost 106490
Execution cost 83618
TABLE VIII. C ASE 1: S TEP 5 D
From 0x14723a09acff6d2a60dcdf7aa4aff308fddc160c
Seller.buyItem(uint256,string,address) C. Case 3: Seller Provides Incorrect Goods.
To
0x692a70d2e424a56d2c6c27aa97d1a86395877b3a
Transaction cost 89622 In this case, a flag is set to indicate that the seller is the
Execution cost 66110 one who causes the transport cancellation. See Table X.
4) Step 4: Seller places a mortgage. See Table IV: D. Case 4: Shipper Fails to Deliver Goods. See Table XI.
5) Step 5: Shipper places a mortgage and agrees to deliver
goods. See Tables (V, VI, VII, and VIII): TABLE XI. C ASE 4
www.ijacsa.thesai.org 558 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
VI. R EMARKS ness of the shipper but also improve the overall performance
of decentralized systems. The mechanism works transparently
In the function solidity diagram, the gas consumption (Fig. across participants. Several real-world scenarios have been
6), we see that the amount of gas in the modality does discussed the feasibility of the proposed multi-sessions in
not pass the transaction and the execution are 300000 and boosting the performance and robustness of the COD systems.
200000, respectively. The amount of gas increases due to the The crucial delivering issue that previous research has not
access to smart contracts as well as the complexity of the yet addressed is sufficiently solved. Our proposed mechanism
methods caused. In the Case Study diagram, e.g. Fig. 7, the ensure the overall efficiency of delivering process. We are
amount of Gas in successful trading scenario shows that stable pleased to announce that a new core component of the buyer
transactions and execution do not exceed 60000 and 40000 and seller’s dilemma. The authors release our sources codes for
respectively. It is important to note that the amount of gas further reproducibility and development. We believe that the
for contract transactions of the participants is very small. In integration of multi-session mechanism, blockchain technology
case of transaction of the failure scenarios, the amount of the and smart contracts will cause significant efficiency across
Gas will be lower than that of in the successful trade. The Gas several disciplines.
will be decided by the smart contract when the execution stops,
ensuring the amount of loss during the transaction process is
insignificant. R EFERENCES
[1] M. Halaweh, “Cash on delivery (cod) as an alternative payment method
VII. C ONCLUSION for e-commerce transactions: Analysis and implications,” International
Journal of Sociotechnology and Knowledge Development (IJSKD),
As we have demonstrated, the integration of multi-session vol. 10, no. 4, pp. 1–12, 2018.
mechanism in any cash on delivery systems is very effective. [2] ——, “Intention to adopt the cash on delivery (cod) payment model for
Our proposed idea is given to not only enhance the effective- e-commerce transactions: An empirical study,” in IFIP International
www.ijacsa.thesai.org 559 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Conference on Computer Information Systems and Industrial Manage- Payment Model for E-commerce Transactions: An Empirical Study,” in
ment. Springer, 2017, pp. 628–637. 16th IFIP International Conference on Computer Information Systems
[3] U. Tandon and R. Kiran, “Study on drivers of online shopping and and Industrial Management (CISIM), ser. Computer Information
significance of cash-on-delivery mode of payment on behavioural Systems and Industrial Management, K. Saeed, W. Homenda, and
intention,” International Journal of Electronic Business, vol. 14, no. 3, R. Chaki, Eds., vol. LNCS-10244. Bialystok, Poland: Springer
pp. 212–237, 2018. International Publishing, Jun. 2017, pp. 628–637, part 7: Various
Aspects of Computer Security. [Online]. Available: https://hal.inria.fr/
[4] J. D. Alie and P. E. Vliek, “International cash-on-delivery system and hal-01656252
method,” Jul. 24 2007, uS Patent 7,249,069.
[13] M. Barkhordari, Z. Nourollah, H. Mashayekhi, Y. Mashayekhi, and
[5] A. Asgaonkar and B. Krishnamachari, “Solving the buyer and seller’s M. S. Ahangar, “Factors influencing adoption of e-payment systems:
dilemma: A dual-deposit escrow smart contract for provably cheat-proof an empirical study on iranian customers,” Information systems and e-
delivery and payment for a digital good without a trusted mediator,” business management, vol. 15, no. 1, pp. 89–116, 2017.
arXiv preprint arXiv:1806.08379, 2018.
[14] “Sites like ebay or etsy but decentralized - our features.” [Online].
[6] H. R. Hasan and K. Salah, “Blockchain-based solution for proof of Available: https://openbazaar.org/features/
delivery of physical assets,” in International Conference on Blockchain.
Springer, 2018, pp. 139–152. [15] N. T. T. Le, Q. N. Nguyen, N. N. Phien, N. Duong-Trung,
T. T. Huynh, T. P. Nguyen, and H. X. Son, “Assuring non-
[7] H. X. Son, M. H. Nguyen, N. N. Phien, H. T. Le, Q. N. Nguyen, fraudulent transactions in cash on delivery by introducing double
V. D. Dinh, P. T. Tru, and P. Nguyen, “Towards a mechanism smart contracts,” International Journal of Advanced Computer
for protecting seller’s interest of cash on delivery by using smart Science and Applications, vol. 10, no. 5, 2019. [Online]. Available:
contract in hyperledger,” International Journal of Advanced Computer http://dx.doi.org/10.14569/IJACSA.2019.0100584
Science and Applications, vol. 10, no. 4, 2019. [Online]. Available:
http://dx.doi.org/10.14569/IJACSA.2019.0100405 [16] G. Wood et al., “Ethereum: A secure decentralised generalised transac-
tion ledger,” Ethereum project yellow paper, vol. 151, pp. 1–32, 2014.
[8] L. J. Camp, J. D. Tygar, and M. R. Harkavy, “Anonymous certified
delivery,” Jun. 13 2000, uS Patent 6,076,078. [17] “Ethereum.” [Online]. Available: https://www.ethereum.org/
[9] N. T. T. Le, Q. N. Nguyen, N. N. Phien, N. Duong-Trung, [18] C. Dannen, Introducing Ethereum and Solidity. Springer, 2017.
T. T. Huynh, T. P. Nguyen, and H. X. Son, “Assuring non- [19] Y. Hirai, “Defining the ethereum virtual machine for interactive theorem
fraudulent transactions in cash on delivery by introducing double provers,” in International Conference on Financial Cryptography and
smart contracts,” International Journal of Advanced Computer Data Security. Springer, 2017, pp. 520–535.
Science and Applications, vol. 10, no. 5, 2019. [Online]. Available: [20] M. Wohrer and U. Zdun, “Smart contracts: security patterns in the
http://dx.doi.org/10.14569/IJACSA.2019.0100584 ethereum ecosystem and solidity,” in 2018 International Workshop on
[10] R. AlTawy, M. ElSheikh, A. M. Youssef, and G. Gong, “Lelantos: A Blockchain Oriented Software Engineering (IWBOSE). IEEE, 2018,
blockchain-based anonymous physical delivery system,” in 2017 15th pp. 2–8.
Annual Conference on Privacy, Security and Trust (PST). IEEE, 2017, [21] “What is a dapp? decentralized application on the blockchain.” [Online].
pp. 15–1509. Available: https://blockchainhub.net/decentralized-applications-dapps/
[11] A. Asgaonkar and B. Krishnamachari, “Solving the buyer and seller’s [22] S. Raval, Decentralized applications: harnessing Bitcoin’s blockchain
dilemma: A dual-deposit escrow smart contract for provably cheat-proof technology. ” O’Reilly Media, Inc.”, 2016.
delivery and payment for a digital good without a trusted mediator,” [23] W. Mougayar, The business blockchain: promise, practice, and appli-
2019 IEEE International Conference on Blockchain and Cryptocurrency cation of the next Internet technology. John Wiley & Sons, 2016.
(ICBC), pp. 262–267, 2018.
[24] A. Wright and P. De Filippi, “Decentralized blockchain technology and
[12] M. Halaweh, “Intention to Adopt the Cash on Delivery (COD) the rise of lex cryptographia,” Available at SSRN 2580664, 2015.
www.ijacsa.thesai.org 560 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Automation has made it possible to garner and mental health and personality traits. However, the application
preserve students’ data and the modern advent in data science of features related to students’ lifestyle and preferences, as
enthusiastically mines this data to predict performance, to the done in this study to predict academic excellence, is a novel
interest of both tutors and tutees. Academic excellence is a approach to the field. In this study, we choose ten such features
phenomenon resulting from a complex set of criteria originating and apply an evidential function—mapping them to students’
in psychology, habits and according to this study, lifestyle and
preferences–justifying machine learning to be ideal in classifying
expertise in the respective field. The study shows that attributes
academic soundness. In this paper, computer science majors’ apart from academic track-records alone can predict academic
data have been gleaned consensually by surveying at Ahsanullah success which can help institutions to foresee the aptitude of
University, situated in Bangladesh. Visually aided exploratory the graduates they are producing, admitting, strategizing for
analysis revealed interesting propensities as features, whose hiring or educating.
significance was further substantiated by statistically inferential
Chi-squared (χ2 ) independence tests and independent samples
Systematic collection of educational data and ML method-
t-tests for categorical and continuous variables respectively, on ologies enable researchers to explore the similarities and dis-
median/mode-imputed data. The initially relaxed p-value retained similarities among academically sound and unsound students.
all exploratorily analyzed features, but gradual rigidification Recent such researches in the EDM arena have gained momen-
exposed the most powerful features by fitting neural networks tum using Neural Networks (NNs). NNs are surpassing tradi-
of decreasing complexity i.e., having 24, 20 and finally 12 hidden tional learning models such as Logistic Regression, Support
neurons. Statistical inference uniquely helped shed off weak Vector Machines in performance—characteristically having
features prior to training, thus optimizing time and generally multiple hidden layers with different activation functions. NNs
large computational power to train expensive predictive models. are versed in fitting complex functions spread through many
The k-fold cross-validated, hyper-parametrically tuned, robust dimensions featuring multiple independent variables. Back-
models performed with average accuracies wavering between
90% to 96% and an average 89.21% F1-score on the optimal
propagation allows refinement of its initial parameters through
model, with the incremental improvement in models proven by numerous epochs, with derivatives showing the direction and
statistical ANOVA. learning rate indicating the magnitude of refinement. The
weights represent a hierarchical mapping from lower (learns
Keywords—Educational Data Mining (EDM); Exploratory Data comparatively simpler features) layers to the higher (learns
Analysis (EDA); median and mode imputation; inferential statistics; sophisticated features) layers.
t-test; Chi-squared independence test; ANOVA-test
The research work addresses a binary classification prob-
lem in categorizing final-year Computer Science (CS) students
I. I NTRODUCTION from Ahsanullah University, Bangladesh as of their academic
The research field of Educational Data Mining (EDM) ap- performance, following the four EDM phases [2]:
plies statistics and machine learning to information stemming • It is generally held that if a CS student is able to
from educational environments and is thus contributing to edu- maintain a CGPA ≥ 3.40 until the final semester,
cational psychology. EDM leverages precise, fine-grained data he/she is faring academically well. First, we explorato-
to discover types of learners, examine effectiveness/suggest rily choose unconventional, unique features by finding
improvements of instructional learning environments, predict their consistent relations with CGPA.
students’ learning behavior and advance learning sciences.
Baker, Yacef [1] critically identified learners, educators, re- • Then the best use of available data is made by imput-
searchers and administrators to be the four stakeholders of ing both categorical and continuous variables.
EDM. • Third, NN models are proposed to predict academic
status.
The bulk of the academic literature, while addressing
problems from the domain of EDM, has taken past aca- • The models and features are statistically cross-
demic credentials into account. Fewer academicians resorted to validated and finer conclusions are drawn.
www.ijacsa.thesai.org 561 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The sequencing of this paper renders the second section as B. Unsupervised Clustering Approaches
a review of existing literature, the third section as descriptions
of methods followed, the fourth section as a depiction of Oyelade et al. [7] analyzed students’ data at a private
experimental results and the final section as concluding notes. Nigerian institution using k-means clustering. The cluster
analysis was combined with standard statistical methods and
a deterministic model was k = 3-fold cross-validated using
II. R ELATED W ORKS different cluster sizes. The study clustered students labeling
them in 5 categories depending on marks’ thresholding. How-
Artificial intelligence-based and statistically analytical ever, the study utilized typical academic indicators. Shovon et
methods (Fig. 1) applied in classifying academic performance al. [8] utilized k-means clustering to analyze learning behavior
can be discussed in light of three prototypical dimensions as in terms of quizzes, mids and finals in three classes.
below.
C. Supervised, Parametric Learning Approaches
Bhardwaj et al. [9] applied a Naive Bayes classifier on the
data of 300 students by preprocessing and transforming the
features of raw data. They selected features with probabilities
> 0.5. They classified among four classes: first, second, third
and fail. The study succeeded in finding interesting features
such as living location, mother’s qualifications etc. Naser et al.
[10] devised an NN based on multilayer perceptron topology
and trained it using sophomores’ data of five consecutive engi-
neering intakes. They considered high school scores, scores at
math and circuitry-based courses during freshman-year, gender
among the predictors—gaining 80% accuracy on test-set.
Fig. 1. Comparison among related researches. Arora et al. [11] proposed a fuzzy probabilistic NN model
for generating personalized prediction which outperformed
traditional ML models. The personalized results showed cross-
stream generalization capabilities and produced 90%, 96%
A. Conventional Statistics and Decision Trees and 87.5% accuracies on three ranks upon training over 570
instances. The model converged to an error of 0.0265 and
Wilkinson, Zhang et al. [3] conducted a study on 706 included interest, belief, family etc. among eighteen features.
undergraduate medical students in three consecutive years at Taylan et al. [12] designed an adaptive neuro-fuzzy inference
the University of Queensland with their objective of modestly system (ANFIS), a combination of NN and fuzzy systems,
determining how precisely each of prior academics, admission to enhance speed and adaptability. The new trend in soft
tests and interviews accounted for students’ performance at computing produced predictions of students’ academics with
post-graduation. These altogether served as the selection crite- crisp numerics. Mueen et al. [13] took into account academic
ria which accounted for 21.9% variation in overall scores. They participation and scores of two courses and modeled them
explored GPA to correlate most strongly with performance (p- to Naive Bayes, NN and decision tree—finding the Bayesian
value < 0.001), followed by interviews (p-value = 0.004) and classifier to provide the highest accuracy of 86%.
admission tests (p-value = 0.08), respectively.
Chamorro-Premuzic et al. [4] established through two III. I MPLEMENTED M ETHODOLOGY
longitudinal studies (sample size, n = 70, 75 respectively)
Ethical collection of students’ data, followed by ex-
that personality-measures could testify for students’ academic
ploratory analysis, preprocessing, predictive modeling and
ability. The setting examined students over three academic
methodical estimation of metrics led to interesting findings
years at two British universities along academic behavior
(Fig. 2).
and personality traits. Sample-1 proved that neuroticism neg-
atively and conscientiousness positively impacted students’
academics, accounting for 10% variance. Sample-2 used EPQ- A. Preparation of AUST CS Students’ Data
R showing three personality factors were instrumental in 1) Collection of Final Semester’s Data:
predicting academic performance and accounted for 17% vari-
ance. • Questionnaire: Students’ responses were gathered via
a survey containing questions of multifarious forms
Yadav et al. [5] explored C4.5, ID3 and CART decision including numerical entries, multiple choices and sen-
trees on engineering students’ data to predict final exam’s tential expressions.
scores. They obtained a true positive rate (TPR) of 0.786
on the ‘fail’ class using ID3 and using C4.5 decision trees, • Environmental setting: The subjects were surveyed
the highest accuracy of 67.77%. Ahmad et al. [6] proved the using Google forms and the responses were recorded
impact of demographic information of students spanning eight as structured data. There were multiple phases of data-
educational years in predicting academic success. They found collection either in the labs of AUST or within the
rule-based classification techniques to fit the data best with comfort of home. No time-constraint allowed subjects
71.3% accuracy. to amply think before responding.
www.ijacsa.thesai.org 562 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE I. I NFERRED STATISTICAL SIGNIFICANCE OF FEATURES 5) Normalization of Input Features: Preprocessing man-
dates inputs and parameters to belong to the same range
Pearson’s χ2 -test
and scale for fair comparison and for the gradient descent to
degrees converge following an aligned orientation.
discrete features χ2 of p-value
freedom
daily hours on FB,
45.254 1 1.73E-11
X − Xmin
state of CGPA Xnew = (3)
classnote-taking tendency, Xmax − Xmin
18.553 2 9.36E-05
state of CGPA
interest in theory, The above formula rescaled all numerics (both categorical:
4.956 2 8.39E-02
state of CGPA
living with family, gender, attendance, interest, etc. and continuous: income, daily
2.7991 1 9.43E-02
state of CGPA hours) within the range [0, 1].
interest in sessional,
2.7272 2 2.56E-01
state of CGPA 6) Maintained Division of Data and k-fold Datasets:
attendance in class,
state of CGPA
1.978 2 3.72E-01 Standard ML practices have been followed by assigning a
gender, larger set of 80% (83 examples) of total examples for training
0.2086 1 6.48E-01
state of CGPA and the rest 20% (20 examples) for cross-validation. The
motive fulfilled motivation,
state of CGPA
0.59718 2 7.42E-01 original distribution of data, i.e., 22.33% positive and 77.67%
Welch Two Sample t-test negative examples, have been maintained throughout training
degrees and test data, in order to eliminate any bias during training or
continious feature t-score of p-value
freedom
cross-validation (Fig. 7).
daily programming hours,
0.21972 36.864 8.27E-01
state of CGPA
monthly income,
-0.63789 24.137 5.30E-01
state of CGPA
1
V ar(W ) = (4)
nin
(n)
X X
−logL({y (n) }, {ŷ (n) }) = [− yi log ŷi ]
n i
X (5)
(n)
= H(y , ŷ (n) )
n
∂
θj := θj − α J(θ) (6)
∂θj
www.ijacsa.thesai.org 566 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
To modify the above for a set of more than one examples, to concoct two bottom-up graphs in order to determine entries
the statement should be replaced by the algorithm below: associated (Fig. 9(a)) with confusion matrices and accuracy on
cross-validation set.
Repeat until convergence {
m After equality-checking, the boolean vector of outputs gave
(i) ‘high’s against the examples identified correctly and ‘low’s
X
θj := θj + α (y (i) − hθ (x(i) ))xj (f or every j) (8)
i=1
against the converse as to having a CGPA above the threshold.
The mean of this data structure rendered the fraction of correct
} identification (Fig. 9(b)).
5) Adam Optimization to Gradient Descent: Adam is a 2) Determination of Metrics from Confusion Matrix: In the
first-order gradient-based optimization algorithm for stochastic domain of statistical classification, a confusion matrix (Fig.
objective functions, using adaptive estimates of lower-order 10(a)) is a special type of contingency table with identical
moments. The parameters used for Adam in this study are as sets of classes in both dimensions—used to account for the
follows: performance of a classification model on cross-validation data
• α : The learning rate or step size, whose decay is for which the actual labels are available.
permissible for Adam, but has not been used.
• β1 : The exponential decay for first-order moment
estimates (e.g. 0.9).
• β2 : The exponential decay for second-order moment
estimates (e.g. 0.999).
• : An infinitesimal number to prevent division by 0
in the implementation (e.g. 10E-8).
C. Estimation of Metrics
1) Creation of Computation Graphs: A computation graph
is a collective mathematical function represented using the
frameworks of graph theory. The round nodes indicate op-
erations while the rectangular ones denote operands, with
the directed edges delineating the sequence of mathematical
operations performed.
Fig. 10. Confusion matrices of our models for some random k-th
cross-validation
Fig. 12. 5-Fold cross-validated results upon training 10, 6-featured 3-layer models (6-featured 3-layer model better fitting the data by overcoming overfitting)
leading us to safely conclude, the models have a systematic The curious problem of predicting students’ perfor-
effect on the accuracy and similar results can be expected if mance has, till date, been addressed using direct predictive
further data-points are added. modeling—this paper proves the effectiveness of visually ex-
ploratory and statistical analysis prior to that objective, leading
A comparative analysis (Fig. 13) reveals that the most to the following landmarks.
www.ijacsa.thesai.org 569 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
• The study avoids random, carefree, holistic selection [5] Surjeet Kumar Yadav and Saurabh Pal. Data mining: A prediction for
of features by first examining their relevance through performance improvement of engineering students using classification.
hypothesis testing, thus establishing the importance of arXiv preprint arXiv:1203.3832, 2012.
statistical preprocessing. [6] Fadhilah Ahmad, Nur Hafieza Ismail, and Azwa Abdul Aziz. The
prediction of students’ academic performance using classification data
• The research endorses data-engineered median and mining techniques. Applied Mathematical Sciences, 9(129):6415–6426,
mode imputation in handling missing values, intro- 2015.
ducing no outside noise to training data. [7] OJ Oyelade, OO Oladipupo, and IC Obagbuwa. Application of k means
clustering algorithm for prediction of students academic performance.
• The paper testifies robustness of the incrementally arXiv preprint arXiv:1002.2425, 2010.
developed proposed models through k-fold cross- [8] Md Hedayetul Islam Shovon and Mahfuza Haque. Prediction of student
validated, ANOVA-tested, significant results. academic performance by an application of k-means clustering algorithm.
International Journal of Advanced Research in Computer Science and
It is recognized that setting the threshold to a CGPA of 3.40 Software Engineering, 2(7), 2012.
may not epitomize aptitude, which depends on factors external [9] Brijesh Kumar Bhardwaj and Saurabh Pal. Data mining: A predic-
tion for performance improvement using classification. arXiv preprint
to the scope of this endeavor. However, this study approves arXiv:1201.3418, 2012.
and incentivizes further researches to consider lifestyle and [10] S Abu Naser, Ihab Zaqout, Mahmoud Abu Ghosh, Rasha Atallah,
personal preferences as useful features towards that end. and Eman Alajrami. Predicting student performance using artificial
neural network: In the faculty of engineering and information technology.
R EFERENCES International Journal of Hybrid Information Technology, 8(2):221–228,
2015.
[1] Ryan SJD Baker and Kalina Yacef. The state of educational data mining
in 2009: A review and future visions. JEDM— Journal of Educational [11] Nidhi Arora and JR Saini. A fuzzy probabilistic neural network for
Data Mining, 1(1):3–17, 2009. student’s academic performance prediction. International Journal of
Innovative Research in Science, Engineering and Technology, 2(9):4425–
[2] Cristóbal Romero and Sebastián Ventura. Educational data mining: a 4432, 2013.
review of the state of the art. IEEE Transactions on Systems, Man, and
Cybernetics, Part C (Applications and Reviews), 40(6):601–618, 2010. [12] Osman Taylan and Bahattin Karagözoğlu. An adaptive neuro-fuzzy
model for prediction of student’s academic performance. Computers &
[3] David Wilkinson, Jianzhen Zhang, Gerard J Byrne, Haida Luke, Ieva Z Industrial Engineering, 57(3):732–741, 2009.
Ozolins, Malcolm H Parker, and Raymond F Peterson. Medical school
selection criteria and the prediction of academic performance. Medical [13] Ahmed Mueen, Bassam Zafar, and Umar Manzoor. Modeling and
journal of australia, 188(6):349–354, 2008. predicting students’ academic performance using data mining techniques.
International Journal of Modern Education and Computer Science,
[4] Tomas Chamorro-Premuzic and Adrian Furnham. Personality predicts 8(11):36, 2016.
academic performance: Evidence from two longitudinal university sam-
ples. Journal of research in personality, 37(4):319–338, 2003. [14] Ng, A., 2000. CS229 Lecture notes. CS229 Lecture notes, 1(1), pp.1-3.
www.ijacsa.thesai.org 570 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Eisa Alanazi
Department of Computer Science
Umm Al-Qura University
Makkah, Saudi Arabia
Abstract—Conditional Preference Networks (CP-nets) are a according to the underlying CP-net? The latter is also known as
compact and natural model to represent conditional qualitative dominance testing, i.e., deciding which alternative dominates
preferences. In CP-nets, the set of variables is fixed in advance. the other. Clearly, the best alternative is the one that is not
That is, the set of alternatives available during the decision dominated by any other alternative or solution. Solving or
process are always the same no matter how long the process is. In answering dominance questions require searching the space
many configuration and interactive problems, it is expected that
some variables are subject to be included or excluded during the
of solutions. Needless to say, for a fixed structure like CP-
configuration process due to users showing interest or boredom nets, the answer to the above questions is the same through
on some aspects of the problem. Representing and reasoning with out the decision process. However, one may expect the answer
such changes is important to the success of the application and to differ from time to time due to some changes happening in
therefore, it is important to have a model capable of dynamically the network.
including or excluding variables. In this work, we introduce active
CP-nets (aCP-nets) as an extension of CP-nets where variable
Also, while having the same answer to both questions
participation is governed by a set of activation requirements. is acceptable on some static domains, it is not the case in
In particular, we introduce an activation status to the CP-net interactive and configuration problems. In the latter, users are
variables and analyze two possible semantics of aCP-nets along usually interested in different subsets of the variables satisfying
with their consistency requirements. certain requirements and hence, the answers need to take into
account the changes. It is intuitive to assume the user interest
Keywords—AI; changes; CP-nets; preferences; decision mak-
ing; product configuration
in having one attribute to be part of the solution space is
conditioned upon the existence of other attributes.
inclusion and exclusion constraints. And, at any given time other outcome that dominates it. It is known that for acyclic
of the decision making process, only those active variables are CP-nets there exists a single optimal outcome that dominates
included in the search. all other solution [3]. For example, the optimal outcome for
CP-net in Fig. 1 is abc. Apparently, the size of the induced
This paper is organized as follows: Background informa- graph is exponential in the number of attributes. Due to the
tion is provided in the next section. Section 3 presents related dependency nature of CP-nets, one needs to consult all the
attempts in the literature. In Section 4, we introduce aCP- values of P a(X) before deciding which value is preferred for
nets and the participation constraints. Section 5 discusses two X. This is mainly because missing a value of any Y ∈ P a(X)
different possible semantics of aCP-nets and show how to solve may lead to inconsistent conclusions, i.e., x being preferred to
dominance testing in each semantic. Lastly, conclusion and x0 and vice versa at the same time. This turns to be the main
foreseeable work is discussed in Section 6. property we need to guarantee when including or excluding
variables and dependencies.
II. C ONDITIONAL P REFERENCE N ETWORKS (CP- NETS )
A CP-net [3] is a graphical model to represent qualitative III. R ELATED W ORK
preference statements including conditional preferences of the Since its inception by Boutilier et al. [3], CP-nets have
form “I prefer x to x0 ” or “I prefer x to x0 when y received a considerable attention from the artificial intelligence
holds”. A CP-net works by exploiting the notion of preferential (AI) community. Many attempts have been proposed tackling
dependency based on the ceteris paribus (with all other things different aspects of CP-nets including their semantics [5]–[8],
being without change) assumption. The CP-net is a set of learning [9], [10], and representation [11], [12]. In particular,
ceteris paribus preference statements which assumed to be several works have been made toward extending the semantics
valid only when two alternatives differ in exactly one variable and the expressive power of the CP-nets. For instance, the work
value. Graphically, a CP-net can be represented by a directed in [7] extended the CP-net to include preference languages
graph where vertices represent features (or variables) V = beyond ceteris paribus and thus allow statements to differ in
{V1 , V2 , . . . , Vn } and arcs represent preference dependencies more than more attribute. Another extension of CP-net is the
among features. Every variable X ∈ V is associated with a weighted CP-net [8] where the user is able to associate weights
set of possible values (its domain) DX . An edge from X to to variables. The work in [12] has introduced (conditional)
Y means the preference of Y depends on the values of X. In importance over variables. Also, [13] extended the model to
such case we say X is a parent of Y and use Pa(Y ) to denote augment the notion of comfort when choosing one alternative
the set of parents for Y . Every variable X is associated with over another. In [14], [15], the preference-based optimization
a ceteris paribus table (denoted as CP T (X)) expressing the problem was investigated where hard constraints are assumed
order ranking of different values of X given the values of the to co-exist with a CP-net and the goal is to find a most
parents P a(X). An outcome for a CP-net is an assignment preferred and feasible solution.
for each variable from its domain. A variable X is an ancestor
of another variable Y if X resides in a path from any root As for extending the semantics of CP-nets to dynamic
node of the graph to Y and X is descendant of Y if Y is an situations, Bigot et al. [4], [9] studied the case where prefer-
ancestor of X. ences are uncertain and a probability distribution is associated
with a statement. The same problem has been also tackled
Given a CP-net, the users usually have some queries about by [16] where dependencies are associated with probability of
the preference statements in the network. One of the main existence and every variable X is associated with a distribution
queries is to find the best outcome given the set of preferences. over the set of total orders for X 0 s values. The work in [11]
We say an outcome o is better than another outcome o0 if considered situations where a webpage content is governed by
there exists a sequence of worsening flips going from o to a CP-net in an adaptive way. Based on the user clicks, the
o0 [3]. A worsening flip is a change in the variable value most preferred content is rendered on the page.
to a less preferred value according to the variable’s CPT.
The relation between different outcomes for a CP-net can be However, none of previous attempts has discussed the
captured through an induced graph. The graph is constructed as dynamic aspect of the CP-nets in handling changes that is
follows: Each node in the induced graph represents an outcome deterministic and of incremental nature. In particular, the vari-
of the network. An edge going from o0 to o exists if there is ables’ inclusion and exclusion during the search. To this end,
an improving flip according to the CPT of one of the variables there are dynamic models to represent configuration problems
in o0 all else being equal. similar to the problems tackled in this work. However, they
target different knowledge information. One notable represen-
Consider the simple CP-net and its induced graph shown in tation in this class is the conditional constraint satisfaction
Fig. 1. The CP-net has three variables A, B and C where A and problem [17] where constraints and variables are included or
B are unconditionally prefer a and b to ā and b̄ respectively. excluded during the search. The conditional CSP formalism is
However, the preference function for C depends on different limited to constraints and cannot directly applied to qualitative
values of A and B. For instance when A = a and B = b̄, the preferences as it is the case in this work.
decision maker prefers c̄ to c as value of the variable C. The
induced graph represents all the information we need to answer A closely related area is the preference-based product con-
different dominance relations between outcomes. An outcome figuration systems [18]–[22] where a configurator is responsi-
o dominates another outcome o0 if there is a path from o0 to o ble for customizing the product based on the user preferences.
in the induced graph otherwise they are incomparable (denoted Such configurators allow for a greater flexibility in meeting
as o ./ o0 ). An outcome is said to be optimal if there exists no the users needs and desires. However, we are not aware of
www.ijacsa.thesai.org 572 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Extending Conditional Preference Networks to Handle Changes Vol. 10, No. 9, 2019
Similarly, Always Required Variable (ARV) is used to look for whether any of X’s descendants will be activated in
include a variable X when subset of variables A1 , ..., An ∈ V Ĉ.
are active regardless of their values and has the form of
incl Definition 4 (Consistent Exclusion). An aCP-net has the
A1 ∧ ... ∧ An −−→ X. consistent exclusion property if for any c ∈ Ĉ whenever a
2) Excluding Variables: Intuitively, Required Not (RN) variable X ∈ V to be excluded, X has no descendants or for
requirement asserts the exclusion of variable based on other any variable Y ∈ V of X’s descendants, there is no c ∈ Ĉsuch
variables values. RN has the form A1 = a1 ∧ A2 = a2 ∧ ... ∧ that Y will be activated.
excl
An = an −−→ X where Ai ∈ V and X ∈ (V − VI ). Similar Definition 5 (Consistent aCP-net). aCP-net is consistent if it
excl
to RN, ARN has the form of A1 ∧ ... ∧ An −−→ X. satisfies the consistent inclusion and exclusion properties.
V. S EMANTICS OF ACP- NETS The goal of conservative rules is to reflect precisely dif-
ferent valid CP-nets from the original CP-net hV, φi without
So far we have described the conditions under which vari- violating its semantics and dependencies. This is formally
ables may be included or excluded from the network domain proved by the following lemma:
without relating them to the underlying aCP-net structure
Lemma 1. If A is a consistent aCP-net then the set of
and semantics. Arbitrary changes might lead to violating the
variables available at any given time of the search represents
semantics of CP-nets. For instance, assume removing one of
a valid CP-net.
the parents of a variable X. How CPT(X) should be updated
for such changes? or consider including a variable X where
one of its parents is not active, how the aCP-net should behave Proof: The proof is by contradiction. Assume A to be a
in such circumstances? Therefore, in this work, we study consistent aCP-net but the set of variables available at time t
different possible changes and define conservative and open form a CP-net that is not valid. By definition, this means there
rules for applying different changes into the aCP-net structure. exists at least one variable X where Y ∈ P a(X) was not
The goal of conservative rules is to represent a valid CP-net included at time t but X was included. First assume X ∈ VI ,
(defined below) at anytime of the aCP-net process. On the this is impossible as VI is available at any given time, and for
other hand, the open semantics aim to represent most general any X, the set of parents must be part of the initial variables
case where the resulted instance of aCP-net are not necessarily as well. Second, assume X 6∈ VI , then there must exist at least
a semantically correct CP-net. one participation constraint c ∈ Ĉ that result in including X.
Given that Y ∈ P a(X) was not included at time t, then the
inclusion was not consistent and thus the aCP-net does not
A. Conservative Semantics have the consistent inclusion property which contradicts with
The core concept here is that the changes must result in a our assumption of A being a consistent aCP-net.
valid CP-net at any given time of the solution process.
Definition 2 (Valid CP-net). Given a set of variables R and Algorithm 1: Consistency Test for aCP-nets
their corresponding CPTs ψ, hR, ψi represents a valid CP-net input : hV, VI , Ĉ, φi: aCP-net Structure
iff for any variable X ∈ R, P a(X) also exists in R. output: True or False
Example 2. Consider R = {A, C, E, F } and ψ is their CPTs 1 foreach c ∈ Ĉ do
for the CP-net in Fig. 2, here R does not represents valid 2 Let X = result in c
CP-net since the variable B ∈ P a(C) is not in R. 3 if c is inclusion condition then
4 foreach Y ∈ P a(X) do
In the conservative semantics of aCP-nets the changes will 5 if STATUS(Y )==INACTIVE then
always result in a valid CP-net. To reflect the conditional de- 6 Return False
pendencies in the structure of CP-net, we assert the activation 7 end
of a variable based on its parents activation. That is for any 8 end
variable X with set of parents P a(X) ⊂ V , for any parent 9 end
variable I ∈ P a(X), either STATUS(I)==ACTIVE or there 10 else
exists c ∈ Ĉ where I will be activated. 11 foreach P ∈ Descendants(X) do
12 if STATUS(P )==ACTIVE then
Definition 3 (Consistent Inclusion). An aCP-net has the 13 Return False
consistent inclusion property if for any c ∈ Ĉ whenever a 14 end
variable X ∈ V to be included, P a(X) is also included in 15 end
the domain. 16 end
17 end
In the context of aCP-net, we need to be careful in 18 Return True
excluding variables. The excluded variable X may be either
a leaf node (thus there are no other variables depend on it)
or not a leaf node in the aCP-net structure. In the first case, Although this might seem too restrictive conditions, it
we can safely remove X since we are guaranteed that there may apply in different domains where the changes are known
will be no other variable S where X ∈ P a(S) that might be a priori and the dependencies between variables cannot be
activated later. In the second case, we can use a procedure to changed. We describe a procedure in Algorithm 1 to check
www.ijacsa.thesai.org 574 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A B ab:cc̄ A B ab:cc̄ A B
ab̄:c̄c ab̄:c̄c
b:dd¯ b:dd¯ b:dd¯
āb:cc̄ āb:cc̄
¯
b̄:dd ¯
b̄:dd ¯
b̄:dd
āb̄:c̄c āb̄:c̄c
D C D C D
c:eē
c̄:ēe
variable to the domain and then reason about its most preferred to abc̄ in order to know the order when A = a. In particular, if
value. abc abc̄ then the CPT(C) is updated to c c̄ for A = a. The
same goes for A = ā with the query whether ābc is preferred
To add dependency between X and Y where X, Y ∈ V . to ābc̄. Such queries hold the promise of revising invalid CPTs
The only condition here is that the new dependency will not and make them valid during the process of decision making.
lead to cycles in the aCP-net structure. We refer to I(X, Y ) as
a dependency of Y on the values of X (i.e., for different x ∈ X
we have an order over DY ). Adding dependency I(X, Y ) will VI. C ONCLUSION AND F UTURE W ORK
result in updating the CPT(Y )∈ φ in a way that P a(Y ) = This paper presented aCP-net, an extension for CP-nets
P a(Y ) ∪ X and for each unique assignments of the parents, to include and exclude variables during the search. We listed
we have an order over DY . some consistency conditions under which the resulted changes
2) Removing Dependencies and Variables: Before remov- always form valid CP-nets and, thus will preserve the seman-
ing a dependency or variable, we first introduce the process of tics of CP-nets. We have also analyzed the situation of changes
marginalization of a variable X ∈ P a(Y ) in CPT(Y ). leading to inconsistencies in the preference information and
suggested possible techniques to overcome the inconsistency
Definition 6 (Marginalization). Given a CPT ` for a variable and answer the dominance testing.
Y , marginalising X ∈ P a(Y ) over ` (denoted as `↓X ) is a
new CPT λ where Dλ =D`−X and for any value x ∈ DX , x Foreseeable work include defining relaxed conditions to
has been removed from λ. allow arbitrary changes over variables and dependencies.
Another important future work is to learn the participation
After marginalising X over CPT(Y ) (CP T (Y )↓X ) it might requirements from historical interactions with the system.
be the case where we have the same assignment of the parents This holds the promise of lowering the burden of specifying
with different orders over DY . Thus, we next provide a participating requirements by the end users.
definition for valid CPTs in the aCP-net structure.
R EFERENCES
Definition 7 (Valid CPT). CPT(X) is a valid CPT iff for each
assignment γ of the parents, we have the same ordering over [1] T. Walsh, “Representing and reasoning with preferences,” AI Magazine,
DX . vol. 28, no. 4, pp. 59–70, 2007.
[2] C. Domshlak, R. I. Brafman, and S. E. Shimony, “Preference-based
configuration of web page content,” in IJCAI, 2001, pp. 1451–1456.
For instance, consider the CP-net in Fig. 1, assume we
[3] C. Boutilier, R. I. Brafman, C. Domshlak, H. H. Hoos, and D. Poole,
are interested in removing the dependency between B and C “Cp-nets: A tool for representing and reasoning with conditional ceteris
(I(B, C)). First we marginalise B over CP T (C). The resulted paribus preference statements,” J. Artif. Intell. Res. (JAIR), vol. 21, pp.
CPT is not valid since for A = a we have two different orders. 135–191, 2004.
[4] D. Bigot, H. Fargier, J. Mengin, and B. Zanuttini, “Probabilistic condi-
Lastly, in order to remove a variable X from the domain, tional preference networks,” in Proc. 29th Conference on Uncertainty
we first need to remove the set of dependencies hold between in Artificial Intelligence (UAI 2013), 2013.
X and its immediate descendants (i.e., children) and then we [5] R. I. Brafman and Y. Dimopoulos, “A new look at the semantics and
can safely remove X. optimization methods of cp-networks,” in IJCAI, 2003, pp. 1033–1038.
[6] C. Boutilier, F. Bacchus, and R. I. Brafman, “Ucp-networks: A directed
3) Posing Queries: Consider removing I(B, C) from the graphical representation of conditional utilities,” in Proceedings of
CP-net structure in Fig. 1, CPT(C) will have the following the Seventeenth conference on Uncertainty in artificial intelligence.
statements: a : c c̄, a : c̄ c, ā : c c̄ and ā : c̄ c. Morgan Kaufmann Publishers Inc., 2001, pp. 56–64.
Obviously, these statements contradict with each other and [7] N. Wilson, “Extending cp-nets with stronger conditional preference
statements,” in AAAI, vol. 4, 2004, pp. 735–741.
breaks the intuitive meaning of CP-net of having exactly one
order for the same assignment of the parents. How the CPT(C) [8] H. Wang, J. Zhang, W. Sun, H. Song, G. Guo, and X. Zhou, “Wcp-
nets: a weighted extension to cp-nets for web service selection,” in
should be updated in such cases? We can overcome such International Conference on Service-Oriented Computing. Springer,
contradictions by revising the order of the variable. One way to 2012, pp. 298–312.
do so is to engage the user by asking different questions. In this [9] D. Bigot, J. Mengin, and B. Zanuttini, “Learning probabilistic cp-nets
particular example, we can ask the user whether she prefers abc from observations of optimal items.” in STAIRS, 2014, pp. 81–90.
www.ijacsa.thesai.org 576 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[10] F. Koriche and B. Zanuttini, “Learning conditional preference net- benefit analysis in product configuration systems,” in Configuration
works,” Artificial Intelligence, vol. 174, no. 11, pp. 685–703, 2010. Workshop 2018 (ConfWS 2018). CEUR-WS, 2018.
[11] R. I. Brafman, “Adaptive rich media presentations via preference-based [22] E. Gençay, P. Schüller, and E. Erdem, “Applications of non-monotonic
constrained optimization,” in Proceedings of the IJCAI-05 Workshop on reasoning to automotive product configuration using answer set pro-
Advances in Preference Handling, 2005. gramming,” Journal of Intelligent Manufacturing, vol. 30, no. 3, pp.
[12] R. I. Brafman, C. Domshlak, and S. E. Shimony, “On graphical 1407–1422, 2019.
modeling of preference and importance,” J. Artif. Intell. Res. (JAIR), [23] M. Sabin, E. C. Freuder, and R. J. Wallace, “Greater efficiency for
vol. 25, pp. 389–424, 2006. conditional constraint satisfaction,” in CP, 2003, pp. 649–663.
[13] S. Ahmed and M. Mouhoub, “Extending conditional preference network
with user’s genuine decisions,” in 2018 IEEE International Conference
on Systems, Man, and Cybernetics (SMC). IEEE, 2018, pp. 4216–4223.
[14] C. Boutilier, R. I. Brafman, C. Domshlak, H. H. Hoos, and D. Poole,
“Preference-based constrained optimization with cp-nets,” Computa- a ā
tional Intelligence, vol. 20, no. 2, pp. 137–157, 2004.
[15] E. Alanazi and M. Mouhoub, “Variable ordering and constraint propa-
gation for constrained cp-nets,” Applied Intelligence, vol. 44, no. 2, pp.
437–448, 2016. b b̄ b b̄
[16] C. Cornelio, J. Goldsmith, N. Mattei, F. Rossi, and K. B. Venable,
“Updates and uncertainty in cp-nets,” in Australasian Joint Conference
on Artificial Intelligence. Springer, 2013, pp. 301–312.
d d¯ d¯ d d d¯
[17] S. Mittal and B. Falkenhainer, “Dynamic constraint satisfaction prob-
lems,” in AAAI, 1990, pp. 25–32.
[18] Y. Wang, D. Y. Mo, and M. M. Tseng, “Relative preference-based
product configurator design,” Procedia CIRP, vol. 83, pp. 575–578, c c̄ c̄ c
2019.
[19] H. L. Jakubovski Filho, T. N. Ferreira, and S. R. Vergilio, “Preference
based multi-objective algorithms applied to the variability testing of f f¯ f¯ f ē e e ē
software product lines,” Journal of Systems and Software, vol. 151, pp.
194–209, 2019.
[20] P. Zheng, Z. Sang, R. Y. Zhong, Y. Liu, C. Liu, K. Mubarok, S. Yu,
X. Xu et al., “Smart manufacturing systems for industry 4.0: Conceptual f¯ f f¯ f f f¯ f f¯
framework, scenarios, and future perspectives,” Frontiers of Mechanical
Engineering, vol. 13, no. 2, pp. 137–150, 2018.
Fig. 4: The complete search space for Example 4
[21] S. Shafiee, A. Felfernig, L. Hvam, P. Piroozfar, and C. Forza, “Cost
www.ijacsa.thesai.org 577 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—In this paper, we successfully designed and de- In this paper, blockchain database is implemented with IoT
veloped Thai agriculture products traceability system using devices.
blockchain and Internet of Things. Blockchain, which is the
distributed database, is used for our proposed traceability system Blockchain is a new technology that catch many attentions
to enhance the transparency and data integrity. OurSQL is in various fields of researches at this moment. The structure of
added on another layer to easier query process of blockchain blockchain illustrates in Fig. 1. It is a chain of blocks which
database, therefore the proposed system is a user-friendly system, each block stores all information of network activities after
which cannot be found in ordinary blockchain database. The the block was added to the chain [6]. This feature makes
website and android application have been developed to show blockchain become traceable database. Blockchain allows ev-
the tracking information of the product. The blockchain database ery user to add data as a transaction. Moreover, all data in the
coupling with Internet of Things give a number of benefits for
our traceability system because all of the collecting information
blockchain can be reviewed by every user, but no one able to
is in real-time and kept in a very secured database. Our system change it [7]. The security and validity of the information in
could have a huge impact on food traceability and supply chain blockchain controlled by one process called 0 mining process0
management become more reliable as well as rebuild public [8]. This process aims to verify the information before adding
awareness in Thailand on food safety and quality control. any block to the chain. A person who verifies the new
transactions and create the block known as miners. Miners
Keywords—Blockchain; internet of things; supply chain man-
utilize a consensus algorithm to add the new block. This
agement; product traceability; distributed database; data integrity; algorithm is similar to a voting system. Once the transaction
ourSQL is confirmed by enough number of nodes (voter), it will be
validated and permanently be a part of the database. After
the block was added to the database, it cannot be changed.
I. I NTRODUCTION Therefore, this method ensures the transparency, trust, and
The Kingdom of Thailand is situated in Southeast Asia. traceability in a system. As this reason, blockchain is very
Thailand is an agricultural country. According to The Trading famous among financial applications. The most well-known
Economics website reported that the Gross Domestic Product is Bitcoin, which is a peer-to-peer electronic cash system [9].
(GDP) from agriculture is the major section that raise the In addition, blockchain has one interesting feature called smart
overall GDP in 2018. Although there is the rapid expansion of contract, which is the digital agreement. This feature is suitable
industrial sectors, agriculture is still a majority [1]. for supply chain management because it can create a deal
between farmers and consumers [7].
However, there are a number of problems and concerns
regarding the quality of Thai agricultural products. Not only Internet of things (IoT) is enabling the connection between
due to the plant diseases and chemical contamination, which machine to machine (M2M) over the Internet [10]. With
is the main problems, but also the uncontrollable factor such IoT, sensors, actuators and electronic devices can collect and
as weather conditions or disasters. These kind of problems exchange the data with each others [11]. In food supply chain
cause the reduction of agricultural products quality. Eventually, traceability system, the IoT devices is calibrated to make sure
this problem leads to the lower income of farmer as well as that its measurement value is accurate. The data collected from
consumer confidence for their receiving products. [1], [2]. the IoT device is reported in real-time [12]. One advantage of
using IoT device to collect the data is because it reading value
The traceability system is an appropriate solution to con- cannot manipulate by human [13]. With the implementation
trol, prevent and exterminate various problems and concerns to blockchain database, all true value is kept directly in the
in supply chain, especially in food and agricultural sector [3]. database without changing, which leads to high reliable of the
In this process the details of the product since it was farming system [14], [15].
until delivered to the consumers has been controlled to meet
required food safety and quality standards (GMP, HACCP, and In this paper, we propose a promising solution in which
HALAL). European Union (EU) has launched a law and policy Internet of Things, blockchain distributed database have been
for the meat products registration and identification to guar- integrated. Section II reviews the works related with tracability
antee that the meats products sold in European countries are system by applying the SQL database, blockchain, and IoT
verifiable and traceable [4]. The traceability system is reliable technology. Section III explains our proposed blockchain trace-
when it consists of secured database and trusted monitoring ability system which consists of the blockchain database tech-
system. Both components can be provided by technology [5]. nology (OurSQL) and system architecture. Section IV presents
www.ijacsa.thesai.org 578 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
A B
Fig. 4. System design and Overview: In the hardware aspect, this system consists of three main parts: Server, IoT device and Smartphone Application (A). For
the service aspect, client service, system management service, and cloud service are three main services need to be separated from each others (B).
user can only see the public information about the product.
2) System management service: It is the control part to
manage the website and send/receive the data between the
blockchain database and IoT devices.
3) Cloud service: In order to show the real-time tracking
system, the Google map API has been used to indicate the
coordinate in the map. With this API feature, we are able Fig. 5. Website illustrates the humidity and temperature and showed as
to track the shipping route in real-time as well as know the virtual gauge (A) and time series graph (B).
specific location of the truck.
stored in ledger. As a result, the sensitive data cannot be stored implementation process. Within this type of blockchain, we
in this kind of the database. Additionally, using a large-scale can assure the security level of the system because every user
network dramatically consumes a lot of time and cost for block has the different priority status. In addition, the smart contract
creation process. is also recommended in order to avoid the middlemen problem.
The deploying of smart contract can be used over many
The implementation of OurSQL, which is a blockchain production steps. For example, we can use the smart contract
database, confirm the information security of our system by to set up the price and the product quality for both farmers
using the proof of work algorithm in order to create the trans- and consumers. Therefore, the farmers are forced to produce
actions and the blocks. Therefore, our proposed methods can their products as good as they can to reach the standard and
work with the sensitive data with reasonable time consuming get the price as mentioned in the contract.
and no block creation fees is needed. This is because OurSQL
is a controllable database technology that will allow only
selected users to use the system and users do not need to pay ACKNOWLEDGMENT
for the block creation fees.
We would like to thank Roman Gelembjuk for the patient
2) Query layer for the data stored in the blockchain: guidance and advice regarding OurSQL integration he has
In order to store the blockchain to hard disk, the data is provided throughout this project development.
required to be compressed. This is meant that these valuable
data sets are extremely hard to be reused or shown to the
users [36]. Moreover, the normal blockchain database is not R EFERENCES
easy to query the information because some specific commands
based on each blockchain technology are required to query [1] N. Poapongsakorn, M. Ruhs, and S. Tangjitwisuth, “Problems and
outlook of agriculture in thailand,” Thailand Development Research
the information [34], [35]. This problem can be solved by Institute Quarterly Review, vol. 13, 01 1998.
implementation of the SQL database with the blockchain
[2] N. Chomchalow, Agricultural development in Thailand. Dordrecht:
technology [7], [36]. In this work, we integrated OurSQL, an Springer Netherlands, 1993, pp. 427–443.
efficiency blockchain database, with the IoT data collection [3] L. U. Opara, “Traceability in agriculture and food supply chain:
method in order to implement the traceability system and A review of basic concepts, technological implications, and future
show the real-time information via both website and android prospects,” in Food Agricultural and Environment, 2003, pp. 101–106.
application. With our proposed database, the website and [4] S. Ammendrup and L. Barcos, “The implementation of traceability
application development can be finished with the use of SQL systems,” Revue scientifique et technique (International Office of Epi-
commands while using a very secured blockchain replication zootics), vol. 25, pp. 763–73, 09 2006.
database as mentioned in subsection III-A. [5] N. V. Vafiadis and T. T. Taefi, “Differentiating blockchain technology to
optimize the processes quality in industry 4.0,” in 2019 IEEE 5th World
Forum on Internet of Things (WF-IoT), April 2019, pp. 864–869.
V. C ONCLUSIONS AND F UTURE W ORK [6] F. X. Tian, “An agri-food supply chain traceability system for china
based on rfid and blockchain technology,” 2016 13th International
In this paper, we have integrated the OurSQL with the Conference on Service Systems and Service Management (ICSSSM),
IoT real-time data collection. To compare with the tradi- pp. 1–6, 2016.
tional system (without the blockchain integration), the use [7] M. Muzammal, Q. Qu, and B. Nasrulin, “Renovating blockchain with
of OurSQL blockchain provides unchangeable data when the distributed databases: An open source system,” Future Generation
data stored into the database. In the aspect of blockchain Computer Systems, vol. 90, pp. 105–117, 2019.
traceability system, our proposed system can utilize the use [8] Feng, Tian, “A supply chain traceability system for food safety based
of SQL commands in the website and application development on haccp, blockchain and internet of things,” in 2017 International
Conference on Service Systems and Service Management, June 2017,
part. Moreover, OurSQL, controllable blockchain database can pp. 1–6.
work with the sensitive data with a faster query time and
[9] S. A. Swamy and N. Jayapandian, “Secure bitcoin transaction and iot
no block creation fees is needed. These are our significant device usage in decentralized application,” in 2018 3rd International
improvement points against the traditional methods both with Conference on Communication and Electronics Systems (ICCES), Oct
and without the blockchain integration. Our system can check 2018, pp. 271–274.
the temperature and humidity of the product in real-time by [10] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, and W. Zhao, “A survey
using website or android application. Moreover, the users can on internet of things: Architecture, enabling technologies, security and
get the notifications when our system found some problems privacy, and applications,” IEEE Internet of Things Journal, vol. 4,
no. 5, pp. 1125–1142, Oct 2017.
related to the temperature and humidity values. In the future,
[11] Y. Yang, L. Wu, G. Yin, L. Li, and H. Zhao, “A survey on security and
we plan to add more sensors in order to get more information privacy issues in internet-of-things,” IEEE Internet of Things Journal,
to ensure the quality of the products. To use all blockchain vol. 4, no. 5, pp. 1250–1258, Oct 2017.
features, future research can further integrating the traceability [12] W. Hong, Y. Cai, Z. Yu, and X. Yu, “An agri-product traceability system
system with Hyperledger blockchain technology, which is a based on iot and blockchain technology,” in 2018 1st IEEE International
permissioned blockchain. This implementation should continue Conference on Hot Information-Centric Networking (HotICN), Aug
using the SQL database in order to allow the user to use SQL 2018, pp. 254–255.
commands with permissioned blockchain. In this permissioned [13] B. Bordel Sánchez, R. Alcarria, D. Martı́n, and T. Robles, “Tf4sm: A
blockchain, the control layer runs on top of the blockchain framework for developing traceability solutions in small manufacturing
companies,” Sensors, vol. 15, pp. 29 478–29 510, 11 2015.
can differentiate the actions that performed by each user.
[14] S. Rahmadika, B. J. Kweka, C. N. Z. Latt, and K. Rhee, “A preliminary
Therefore, permissioned blockchain has a better transaction approach of blockchain technology in supply chain system,” in 2018
performance because we can set the block size limitation and IEEE International Conference on Data Mining Workshops (ICDMW),
the validate information by adjusting the Chaincode during Nov 2018, pp. 156–160.
www.ijacsa.thesai.org 582 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[15] S. Aich, S. Chakraborty, M. Sain, H. Lee, and H. Kim, “A review on in 2018 IEEE/ACIS 17th International Conference on Computer and
benefits of iot integrated blockchain based supply chain management Information Science (ICIS), June 2018, pp. 180–184.
implementations across different sectors with case study,” in 2019 [26] A. H. Ngu, M. Gutierrez, V. Metsis, S. Nepal, and Q. Z. Sheng,
21st International Conference on Advanced Communication Technology “Iot middleware: A survey on issues and enabling technologies,” IEEE
(ICACT), Feb 2019, pp. 138–141. Internet of Things Journal, vol. 4, no. 1, pp. 1–20, Feb 2017.
[16] T. M. Bhatt and J. Zhang, “Food product tracing technology capabilities [27] S. Thattapon and H. Scott C.-H., “Enhancing voip security and effi-
and interoperability.” Journal of food science, vol. 78 Suppl 2, pp. B28– ciency using vpn,” in 2019 International Conference on Computing,
33, 2013. Networking and Communications (ICNC), Feb 2019, pp. 180–184.
[17] B. Adam, R. B. Holcomb, M. Buserc, B. Mayfieldd, J. Thomase, C. A. [28] M. Samaniego and R. Deters, “Blockchain as a service for iot,” in
O’Bryanf, P. Crandallg, D. K. R. Knipei, and S. C. Ricke, “Enhancing 2016 IEEE International Conference on Internet of Things (iThings) and
food safety , product quality , and value-added in food supply chains IEEE Green Computing and Communications (GreenCom) and IEEE
using whole-chain traceability,” in International Food and Agribusiness Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data
Management Review, 2016. (SmartData), 2016, pp. 433–436.
[18] H. Hejazi, H. Rajab, T. Cinkler, and L. Lengyel, “Survey of platforms [29] M. Kim, B. Hilton, Z. Burks, and J. Reyes, “Integrating blockchain,
for massive iot,” in 2018 IEEE International Conference on Future IoT smart contract-tokens, and iot to design a food traceability solution,” in
Technologies (Future IoT), Jan 2018, pp. 1–8. 2018 IEEE 9th Annual Information Technology, Electronics and Mobile
[19] L. B. Campos and C. E. Cugnasca, “Towards an iot-based architecture Communication Conference (IEMCON), Nov 2018, pp. 335–340.
for wine traceability,” in 2015 International Conference on Distributed [30] M. Singh, A. Singh, and S. Kim, “Blockchain: A game changer for
Computing in Sensor Systems, June 2015, pp. 212–213. securing iot data,” in 2018 IEEE 4th World Forum on Internet of Things
[20] W. Liang, J. Cao, Y. Fan, K. Zhu, and Q. Dai, “Modeling and im- (WF-IoT), 2018, pp. 51–55.
plementation of cattle/beef supply chain traceability using a distributed [31] A. Reyna, C. Martı́n, J. Chen, E. Soler, and M. Dı́az, “On blockchain
rfid-based framework in china,” PloS one, vol. 10, p. e0139558, 10 and its integration with iot. challenges and opportunities,” Future
2015. Generation Computer Systems, vol. 88, pp. 173–190, 2018.
[21] K. Wongpatikaseree, P. Kanka, and A. Ratikan, “Developing smart farm [32] K. Salah, N. Nizamuddin, R. Jayaraman, and M. Omar, “Blockchain-
and traceability system for agricultural products using iot technology,” based soybean traceability in agricultural supply chain,” IEEE Access,
in 2018 IEEE/ACIS 17th International Conference on Computer and vol. 7, pp. 73 295–73 305, 2019.
Information Science (ICIS), June 2018, pp. 180–184.
[33] M. P. Caro, M. S. Ali, M. Vecchio, and R. Giaffreda, “Blockchain-based
[22] W. Cao, L. Zheng, H. Zhu, and P. Wu, “General framework for animal traceability in agri-food supply chain management: A practical imple-
food safety traceability using gs1 and rfid,” in CCTA, 2009. mentation,” in 2018 IoT Vertical and Topical Summit on Agriculture -
[23] G. Angel and A. Brindha, “Real-time monitoring of gps-tracking mul- Tuscany (IOT Tuscany), May 2018, pp. 1–4.
tifunctional vehicle path control and data acquisition based on zigbee [34] R. Adams, G. Parry, P. Godsiff, and P. Ward, “The future of money
multi-hop mesh network,” in 2011 International Conference on Recent and further applications of the blockchain,” Strategic Change, vol. 26,
Advancements in Electrical, Electronics and Control Engineering, 2011, no. 5, pp. 417–422, 2017.
pp. 398–400. [35] E. Karafiloski and A. Mishev, “Blockchain solutions for big data
[24] M. U. H. A. Rasyid, E. M. Kusumaningtyas, and F. Setiawan, “Appli- challenges: A literature review,” in IEEE EUROCON 2017 -17th Inter-
cation to determine water volume for agriculture based on temperature national Conference on Smart Technologies, July 2017, pp. 763–768.
amp; humidity using wireless sensor network,” in 2016 International [36] Y. Li, K. Zheng, Y. Yan, Q. Liu, and X. Zhou, “Etherql: A query layer
Conference on Knowledge Creation and Intelligent Computing (KCIC), for blockchain system,” in Database Systems for Advanced Applica-
2016, pp. 105–112. tions, S. Candan, L. Chen, T. B. Pedersen, L. Chang, and W. Hua, Eds.
[25] K. Wongpatikaseree, P. Kanka, and A. Ratikan, “Developing smart farm Cham: Springer International Publishing, 2017, pp. 556–567.
and traceability system for agricultural products using iot technology,”
www.ijacsa.thesai.org 583 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—Electronic cash (or e-cash) research has been going identification successfully discovers the identity of double-
on for more than three decades since it was first proposed. Various spender, but it cannot prevent double-spending in advance,
schemes and methods are proposed to improve privacy and secu- especially for offline e-cash scheme.
rity in e-cash, but there is one security issue that less discussed
mainly in offline e-cash, namely, double-spending. Generally, the Several existing proposed methods to fulfill the property of
mechanism to deal with double-spending in offline e-cash is double-spending prevention such as blockchain [8], smartcard
performing double-spending identification when depositing the [9], and mobile agent [10]. Nakamoto proposed the use of
coin. Even though the mechanism is successful in identifying blockchain in a peer-to-peer e-cash system to prevent double-
double-spender, but it cannot prevent double-spending. This spending [8]. Blockchain is a global ledger where all trans-
paper proposes the Mobile Agent Platform based Wallet (MAPW) action recorded. Every transaction must be broadcasted to
to overcome the double-spending issue in offline e-cash. MAPW
uses the autonomy and cooperation of agents to give protection
all nodes and added to the ledger. Because all nodes keep
against malicious agent, counterfeit coin and duplicate coin. This this ledger, it is impossible to perform double-spending. The
model has been verified using Colored Petri Nets (CPN) and blockchain-based method is comparatively slow in transaction
has proven to be successful in preventing double-spending, and speed and confirmation. As reported by Statista, average con-
overcoming malicious agent, and counterfeit coins. firmation time of bitcoin, which is one of e-cash scheme that
implements the blockchain method, is 9.47 minutes1 .
Keywords—e-Cash; double-spending; MAPW; CPN
In order to manage the double-spending problem in offline
e-cash scheme, Liu proposed a method that uses a smartcard
I. I NTRODUCTION that records a pair of all withdrawn coins [9]. When a customer
spends a coin, a merchant requests the pair of the spent coin
Nowadays, electronic payment has proliferated along with to the smartcard. Smartcard searches the pair to prove that the
the use of the Internet. The electronic payments can be classi- coin has not been spent yet. If the pair exists in the smartcard,
fied into four categories: online credit card, electronic check, merchant accepts the coin while smartcard deletes the pair of
smart cards based electronic payment, and e-cash [1]. E-cash the coin. The customer cannot spend the same coin because
is the only electronic payment that provides not only security a pair of the coin no longer exist in the smartcard. However,
but also the privacy of its users. E-cash generally consists of this method does not provide any mechanism to prevent a pair
three types of entities: bank, user, and merchant. The user of the spent coin rewritten in the smartcard.
withdraws coins from the bank and spends it on the merchant
who then deposits the coins to the bank. The security aspect of Furthermore, Salama proposed a more advanced method in
e-cash should cover some main properties (1) unforgeability: against the double-spending problem by using Optical Memory
no user can create coin other than the authorized party; (2) no Card (OMC) and mobile agent [10]. OMC is a write-only card
framing: no one except the owner of a coin can spend it; and that used for recording the serial number of the spent coin. The
(3) double-spending prevention: coin can only be spent once. mobile agent is used as a coin that can identify the spent coin
The privacy aspect covers the anonymity of users, which mean which its serial number has been recorded in OMC. Hence,
no one can discover the true identity of a user correlated with customer cannot spend the same coin if the serial number of
withdrawal or spending transaction. the coin recorded in OMC. This method can prevent double-
spending in advance but has limitation in OMC memory. When
Following the first e-cash scheme introduced by Chaum OMC has no memory space, the double-spending prevention
[2], [3], many e-cash schemes [4], [5], [6], [7] have been capability cannot be performed.
proposed, and most of them focus on improving privacy and se-
From the above analysis, the existing methods have not
curity (unforgeability and no framing). These proposed e-cash
been able to prevent double-spending optimally. These meth-
schemes only provide double-spending identification to over-
ods still have open issues in the slow confirmation time, the
come double-spending. Double-spending is a security issue in
inability to counter data of spent coin to be copied on the
which the same coin can be spent more than once since an e-
smartcard and limited data storage space on OMC. Double-
cash’s coin is a set of digital data that can be duplicated easily.
spending causes financial losses, so this issue is paramount to
Double-spending identification is a mechanism to identify
whether a coin is a duplicated or not. If the coin is identified as 1 Average confirmation time of Bitcoin transactions from June 2017 to
duplicated coin, e-cash system revoke anonymity and discover June 2018 according to https://www.statista.com/statistics/793539/bitcoin-
the identity of the duplicated coin owner. The double-spending transaction-confirmation-time / accessed 31 July 2019
www.ijacsa.thesai.org 584 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
resolve. Thus, we need to construct a method that meets the Besides, e-cash also must be resistant to counterfeit coin and
double-spending prevention property and prove the security of adversary users.
the method. This paper proposes the Mobile Agent Platform
based Wallet (MAPW) model that is not only able to deter
counterfeit coin and prevent double-spending but also protect Bank
coin coin
the mobile agent platform from malicious agents. withdrawal deposit
II. P RELIMINARIES
Bank
This section introduces some concepts related to offline coin double-spending coin
e-cash system and the use of agent’s technology in the e-cash. withdrawal checking deposit
the customer spends the coin by sending it to a merchant (b) Offline e-cash
in trading for some goods or services (payment protocol).
Finally, the merchant ends the cycle by depositing coin (deposit Fig. 2. Coin verification in traditional e-cash
protocol). There are two types of e-cash schemes, namely,
online and offline.
and unauthorized access. Protection of mobile agent platform 3) The deposit checking: The merchant M can deposit
from a malicious agent can adopt various techniques. The coin ci to the bank B. Before depositing ci , M and B
first is sandbox technique which isolates untrusted agent so perform challenge-response. B allows M to send ci if they
cannot alter the platform or agent in it [19]. Simple Malicious pass the challenge-response. First, B verifies the signature of
Identification Police (MIP) model [20] is the second technique ci and checks whether ci has been previously deposited. If the
that can be adopted to protect the mobile agent platform. The signature of ci is valid and ci is never deposited before, B
concept of this technique is identifying malicious agent by accepts ci .
scanning the byte code of the agent.
4) The synchronization checking: The purpose of the syn-
chronization checking is preventing the wallet from receiving
III. P ROPOSED M ODEL duplicate coin. The synchronization is performed when the
wallet of C or M receives or sends a coin. The wallet checks
The proposed model is the MAPW model for preventing the identity of the new coin whenever it receives a new coin.
double-spending in offline e-cash scheme. This section gives The new coin is accepted if there are no coins in the wallet that
an overview and description of the proposed model in detail. has the same identity as the new coin. Otherwise, the wallet
refuses the new coin. Before the wallet sends a coin, the wallet
A. Overview of Proposed Model checks the existence of coin’s identity. If coin’s identity exists,
it allows the coin to migrate to another wallet.
The MAPW model is intentionally designed as a coin’s
wallet with protection against malicious agent, counterfeit B. Model Description
coin, and double-spent coin. Simple MIP model is adopted to
protect MAPW against malicious agent. In order to overcome MAPW model, as shown in Fig. 4, has four static agents
the counterfeit and double-spent coin, MAPW applies the (like user, bank, identifier, and killer agent) and one mobile
autonomy and cooperation capabilities of the software agent. agent (coin agent). The static agents, with their respective
duties and responsibilities, protect the mobile agent platform,
The main idea of MAPW model is to append double- and ensuring no counterfeit and double-spent coin. User agent
spending identification when receiving a coin thus can prevent performs three e-cash protocols (withdrawal, payment, and
double-spending in advance. Fig. 3 illustrates the proposed deposit), and is responsible for incoming and outgoing check-
offline e-cash cycle used by MPAW. There are three parties: ing. Bank agent is an agent of the bank’s representative that
bank B that able to issue coins and accept deposited coins; stores the identity of all coin agent in the wallet and identifies
customer C that can withdraw and spend coins; and merchant a duplicate agent. Identifier agent determines whether the
M that can accept spending coins and deposit coin. Our foreign agent is a malicious agent or not. Killer agent kills
proposed model is composed of withdrawal protocol, payment any malicious agent, counterfeit coin, and double-spent coin.
protocol, and deposit protocol.
incoming outgoing
request request
double-spending
checking ? incoming coin
agent
Bank incoming checking
coin coin gate
invalid incoming outgoing
withdrawal deposit not coin
checking checking
valid
Customer Merchant
duplication
kills coin coin storage
• coin verification yes checking
MAPW • double-spending prevention MAPW exist
no
remove
add incoming coin ID
outgoing
coin coin ID existence outgoing
coin ID
payment gate
not exist
of the bank and other coin agents that are in the same wallet wallet. The coin that accepts this request then broadcasts its
at the same time. migration to the bank agent and other coins. They delete the
coin’s ID from their memory and allow the coin’s migration
serial
signature origin sync ID
if the coin’s ID is in their memory. Otherwise, they consider
number the coin as an invalid coin and trigger the killer agent to kill
the coin.
Fig. 5. The block data of coin.
Algorithm 2 Algorithm for the leaving of a coin
if request is a valid outgoing coin request then
the user agent sends a request to the coin for migrating
C. The Function of Mobile Agent Platform based Wallet Model to another wallet;
The functional process algorithm of MAPW model in- the coin broadcasts its migration;
volves two algorithms: the arrival and leaving of a coin that if the bank agent and other coins know the ID of the coin
is respectively given in Algorithm 1 and Algorithm 2. In then
Algorithm 1, the arrival of a new agent triggers the identifier other coins agent and bank agent delete ID of the coin;
agent to identify the new agent. If the new agent is not the coin migrates and deletes the ID of other coins;
identified as a coin agent, the identifier agent will trigger the else
killer agent to kill the new agent. Otherwise, the new agent killer agent kills the coin agent and exit;
will be considered as a new coin and forwarded to the user end if
agent for verifying the new coin’s signature. The new coin is else
allowed to broadcast its arrival to the bank agent and stored ignores request;
coins if its signature is valid, but the killer agent will kill it if end if
its signature is invalid. After the bank agent and stored coins
receive the broadcast message, they check the existence of the
new coin ID in their memory of the stored coins’ ID and send IV. C OLORED P ETRI N ETS M ODEL OF M OBILE AGENT
a kill command to the killer agent if its ID is a duplicate. P LATFORM BASED WALLET
However, the bank agent and stored coins save the new coin MAPW is the proposed model of double-spending preven-
ID if the new coin is not a duplicate. The new coin also saves tion in offline e-cash. In order to determine the correctness
the bank agent and all stored coins’ ID. and eliminating or minimizing the security of MAPW, it must
be verified by using a formal method. There are various
Algorithm 1 Algorithm for the arrival of agent/coin formal methods, but the most commonly used for the agent
if the agent is not a coin or request is not a valid incoming is Petri nets. For example, Petri nets can be used for modeling
request then interaction protocol in multiagent system [21] and for verifying
killer agent kills the agent and exit; agent-based architecture [22]. This paper uses CPN, that is
else a combination of the capabilities of Petri nets and a high-
the agent is considered as a new coin; level programming language, for the design, development, and
user agent verifies the new coin’s signature; analysis of MAPW [23].
if the signature of the new coin is invalid then
sends a command to killer agent to kill the new coin; TABLE I. T ESTED SCENARIOS FOR MAPW MODEL
else
the new coin broadcasts its arrival to stored coin and Condition of coin
Agent
bank agent; Case
type Signature
Legitimate
Duplicate
Known
origin ID
the bank agent and stored coins check the new coin ID malicious
whether its already exist or not; not coin - - - -
agent
if the ID of new coin is a duplicate of stored ID coin counterfeit
coin
invalid
- - -
coin (sign=0)
then valid invalid
the bank agent or stored coins send a kill command - -
(sign=1) (orig=0)
to killer agent; double
coin
valid valid
yes -
spending (sign=1) (orig=1)
else valid valid
no no
the bank agent and stored coins save the new coin (sign=1) (orig=1)
ID; normal valid valid
coin no yes
spending (sign=1) (orig=1)
the new coin ID saves all stored coin ID and bank
agent;
end if Table I shows a set of the tested scenario for MAPW model
end if in proving the MAPW’s protection against malicious agent,
end if counterfeit coin, and double-spending. MAPW also should
able to do normal spending. A malicious agent is an agent
that its type is not a coin agent. A counterfeit coin is a coin
The leaving of a coin, as described in Algorithm 2, begins with an invalid signature (sign=0). There are three possibilities
whenever the user agent accepts a valid outgoing request. The of double-spending. First, a coin with a valid signature but
user agent sends a request to a coin for migrating to another not came from the legitimate origin (valid withdrawal or valid
www.ijacsa.thesai.org 587 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
payment). Second, coin with valid signature and came from value from 1 to 500. The boolV al() function returns a random
legitimate origin but has a duplicate in the wallet. Third, a number of 0 or 1. The serialLabel() function returns a random
coin with a valid signature, came from legitimate and does not serial number of a coin.
have a duplicate in the wallet, but its identity is not recognized.
The last scenario is normal spending that is a coin with a valid fun count(synCoin:SYN)=
signature, came from the legitimate origin, does not have any if synCoin=[] then 0
else 1+count(tl(synCoin));
duplicate, and its identity is recognized by bank agent and coin fun notin(sn:SN,syn:SYN)=
agent. if syn=[] then true
else if sn=hd(syn) then false
The CPN model of MAPW consists of one main MAPW else notin(sn,tl(syn));
fun serialVal()=
model page and four subpages for coin’s generation, incoming discrete(1,500);
coin, outgoing coin, and synchronization coin’s ID. Fig. 6 fun boolVal()=
discrete(0,1);
illustrates the main MAPW model page in which accepts the fun serialLabel()=
incoming and outgoing request. Every time MAPW accepts "serial"^Int.toString(serialVal());
incoming coin request, it will trigger coin generation to gen-
erate random coin, and this random coin will be checked by Fig. 8. Declaration of CPN model function
incoming coin checking. If the request is an outgoing coin
request, it will be checked by outgoing coin checking.
CoinGenerate
all coin
synchronization
of the identifier agent that is responsible for checking all
CoinSync
coin
incoming
done
COIN incoming agent. This subpage performs the generation of a
coin
start CoinSync
outgoing
accepted
random incoming agent (coin and non-coin agent) that enters
incoming
coin
entrance
coin
accepted
"bank" MAPW through the incoming gate, which is triggered by gen
coin ID
COIN COIN sync trigger place. Coin agent is represented by 1, while a non-coin
SN
agent is represented by 0. Random value 0 or 1 is generated
by boolV al() function. If a non-coin agent enters MAPW, it
CoinIncomingChecking
store
COIN
CoinOutgoingChecking
will be killed, and the model will return to wait for a request.
coin
1 {serial="bank",sign=1,
However, if the incoming agent is a coin agent, it will generate
CoinIncomingChecking
agentType
coin
orig=1,syn=["bank"]} CoinOutgoingChecking
a coin agent and send the coin agent to coin entrance.
number
INT
agent
transaction kill
type
rejected coin coin INT
check
gen gen 1`boolVal() foreign
INT COIN trigger agent agent
In agentType
Fig. 6. Main CPN model of MAPW
agentType agent
filter
not coin type
if agentType=0 then 1`() foreign agent
check
else empty Out
INT
if agentType=1 then
1`{serial=serialLabel(),
There are four customized color types and five customized kill
foreign agent
sign=boolVal(),
orig=boolVal(),
functions in the CPN model of MAPW. Fig. 7 shows the dec- syn=[]}
else empty
coin
all coin
synchronization
TABLE III. T HE R ESULT OF T ESTED S CENARIOS FOR MAPW M ODEL
outgoing
start done
[#serial(coin)<>"bank"] In Out
transaction
COIN
coin
call coin
coin leaving
if coinNum=1
rejected Scenario Result
start
then 1`()
else empty
Out malicious agent detected and killed
coin
if coinNum>1
then 1`()
if notsafeCnt>0
then 1`coin
counterfeit coin detected and killed
COIN
coin
else empty
coin
else empty double-spending detected and killed
start scan outgoing
COIN
outgoing
checking accepted if notsafeCnt>0 normal spending pass
coinNum then empty
coin Out else 1`coin
1
store start next
coin scan coin
In/Out
COIN coin
coin sync
COIN
prevention, but they do not have any wallet protection. In
starting point 1
coin
order to perform double-spending prevention, Liu [9] uses
if count(serialList)<(coinNum-2)
coin
then 1`coin
else empty In/Out
number
INT
[count(serialList)
=coinNum-1] smartcard and Salama [10] uses OMC and mobile agent. The
coinNum
coinOld coinNum MAPW model has both forgery and double-spending pre-
coin sync coin
sync check
if count(serialList) >= (coinNum-2)
then 1`coin else empty
end point
decide sync result
vention. Double-spending prevention is performed by mobile
[notin(#serial(coinOld),
COIN
agent without depending on the specific hardware. The MAPW
serialList)]
safeCnt 0
model also has wallet protection that protects the wallet from
safeCnt
safe
counter 0
malicious agents.
if notin(#serial(coin),#syn(coinOld)) INT
then 0
else safeCnt+1
notsafeCnt 0 notsafeCnt
www.ijacsa.thesai.org 590 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
[4] S. Canard and A. Gouget, “Anonymity in transferable e-cash,” in Ap- [14] C. Wang, H. Sun, H. Zhang, and Z. Jin, “An improved off-line
plied Cryptography and Network Security, S. M. Bellovin, R. Gennaro, electronic cash scheme,” in International Conference on Computational
A. Keromytis, and M. Yung, Eds. Berlin, Heidelberg: Springer Berlin and Information Sciences. IEEE, 2013, pp. 438–441.
Heidelberg, 2008, pp. 207–223. [15] F. U. Ogban and U. Udoh, “A mobile agent-based distributed infor-
[5] C. I. Fan and V. S. M. Huang, “Provably secure integrated on/off-line mation retrieval system,” International Journal of Natural and Applied
electronic cash for flexible and efficient payment,” IEEE Transactions Sciences, vol. 10, pp. 72–77, 01 2015.
on Systems, Man and Cybernetics, Part C: Applications and Reviews, [16] G. Liu, “The application of intelligent agents in libraries: a survey,”
vol. 40, no. 5, pp. 567–579, September 2010. Program: Electronic Library & Information Systems, vol. 45, no. 1, pp.
[6] O. Blazy, S. Canard, G. Fuchsbauer, A. Gouget, H. Sibert, and J. Traore, 78–97, 2011.
“Achieving optimal anonymity in transferable e-cash with a judge,” [17] S. U. Guan, S. L. Tan, and F. Hua, “A modularized electronic payment
AFRICACRYPT, pp. 206–223, July 2011. system for agent-based e-commerce,” Journal of Research and Practice
[7] J. Zhang, H. Guo, Z. Li, and C. Xu, “Transferable conditional e- in Information Technology, vol. 36, no. 2, pp. 67–87, May 2004.
cash with optimal anonymity in the standard model,” IET Information [18] C. Anhalt and S. Kirn, “Towards payment systems for mobile agents,”
Security, vol. 9, no. 1, pp. 59–72, December 2015. in Proceedings of the 4th European Workshop on Multi-Agent Systems,
[8] S. Nakamoto, “A peer-to-peer electronic cash system,” B. Dunin-Keplicz, A. Omicini, and J. Padget, Eds., vol. 223. CEUR,
http://www.bitcoin.org/bitcoin.pdf, 2009. December 2006.
[9] W. Y. Liu, Y. A. Luo, and Y. L. Si, “A security multi-bank e-cash [19] R. Wahbe, S. Lucco, T. Anderson, and S. Graham, “Efficient software-
protocol based on smart card,” in Proceedings of the Sixth International based fault isolation,” ACM SIGOPS Operationg Systems Review,
Conference on Machine Learning and Cybernetics. IEEE, August vol. 27, no. 5, pp. 203–216, 1993.
2007, pp. 3244–3248. [20] S. Venkatesan and C. Chellappan, “Protection of mobile agent platform
[10] M. A. Salama, N. El-Bendary, and A. E. Hassanien, “Towards secure through attack identification scanner (ais) by malicious identification
mobile agent based e-cash system,” in Proceedings of the First Inter- police (mip),” in 2008 First International Conference on Emerging
national Workshop on Security and Privacy Preserving in e-Societies, Trends in Engineering and Technology, July 2008, pp. 1228–1231.
New York, 2011, pp. 1–6. [21] B. Marzougui and K. Barkaoui, “Interaction protocols in multi-agent
[11] S. H. Islam, R. Amin, G. P. Biswas, M. S. Obaidat, and M. K. Kan, systems based on agent petri nets model,” International Journal of
“Provably secure pairing-free identity-based partially blind signature Advanced Computer Science and Applications, vol. 4, no. 7, pp. 166–
scheme and its application in online e-cash system,” Arabian Journal 173, 2013.
for Science and Engineering, vol. 41, no. 8, pp. 3163–3176, August [22] N. A. Mian and F. Ahmad, “Agent based architecture for modeling and
2016. analysis of self adaptive systems using formal methods,” International
[12] X. Zhou, “Threshold cryptosystem based fair off-line e-cash,” in Second Journal of Advanced Computer Science and Applications, vol. 9, no. 1,
International Symposium on Intelligent Information Technology Appli- pp. 563–567, 2018.
cation, vol. 3. Shanghai: IEEE, 2008, pp. 692–696. [23] K. Jensen and L. M. Kristensen, Coloured Petri Nets: Modelling
[13] W.-S. Juang, “An efficient and practical fair buyer-anonymity exchange and Validation of Concurrent Systems, 1st ed. Springer Publishing
scheme using bilinear pairing,” in 2013 Eight Asia Joint Conference on Company, Incorporated, 2009.
Information Security, 2013, pp. 19–26.
www.ijacsa.thesai.org 591 | P a g e
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The architecture of Next Generation of networks explains the need to reduce the degradation time to ensure the
(NGN) aims to diversify the offer of operators in added value satisfaction of end customers.
services. To do this, NGN offers a heterogeneous architecture for
the services deployment. This poses significant challenges in In this work, we propose a proactive approach to end-to-
terms of end-to-end assurance of services. For this purpose, we end and real-time assurance of services; an approach that will
propose in this work the establishment of a proactive predicts any degradation before sufficient time for correction.
autonomous system, capable of ensuring an acceptable quality To do this we propose a prediction model able to predict the
level according to Service Level Agreement (SLA) requirements. variation of the state of the network and thus to deduce the
A system that is able to predict any QoS degradation due to the possible impairments of the QoS, in real time.
prediction model based on time series adapted to NGN.
This document is organized as follows. At the beginning
Keywords—Next Generation of Network (NGN); network we will present the NGN architecture and the SLM & M [10]
management; enhanced Telecom Operation Management (eTOM) solution which is used for the correction of the degradations.
frameworks; prediction; time series; Ip Multimedia Subsystem Then we will present the new prediction approach and discuss
(IMS); Service Level Agreement (SLA); Quality Of Service (QoS) the model adapted to the NGN context. In the last section of
the document we will discuss the results obtained during the
I. INTRODUCTION experimentation phase in real cases of service provision.
Next Generation of Networks (NGN) [1] offers a standard
II. BACKGROUND AND RELATED WORKS
architecture for easy integration of services with existing
communication technologies. An architecture that aims A. NGN Networks
primarily to facilitate the deployment and provision of value- 3GPP specifications dedicated to NGN networks offers a
added services to customers of telecommunications operators simplified architecture for the core network known by the IP
[2][3]. This will help to evolve the business of the operator Multimedia Subsystem (IMS) architecture [2]. This
from a simple access provider to an end-to-end service architecture aims to simplify the provision of service regardless
provider. of the communication technology used for service consumer
The 3GPP [4] specifications describing the architecture of access. To do this, the IMS offers unified procedures for user
NGNs cut the network in three levels. The access level which authentication and access to services. What is achievable by
groups together existing communication technologies. The cutting the architecture at three levels is as follows:
responsible control level unifies access to networks and
Access Level: This level is responsible for ensuring the
services regardless of the underlying technology and the level
connectivity of users to the control entities described in
of service responsible of the unfolding and provision of value-
the 3GPP specifications. It allows to interface
added services.
effectively with different types of communication
The diversity of communication technologies and the technologies via interface components usually installed
multitude of integrated entities pose significant challenges for in the IMS core boundary routers. The access networks
end-to-end assurance of services [5]. To this end, several are connected to a global network linking the different
studies have dealt with the management of QoS in NGN control entities and the interfacing components.
networks [6][7][8]. However, most of this work proposes
solutions to correct the level of the QoS after a possible The Control Level: This is the key level of the NGN
degradation detected. This significantly affects the quality of architecture; it ensures the control of users and services
the QoE user experience. Indeed, the time required for deployed by the operator. The level basically includes
detecting and correcting degradations is very important, given four entities: The three control entities of the P_CSCF,
the difficulty of real-time correction operations [9]. This I_CSCF and S_CSCF sessions and the HSS database.
592 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Service Level: The level that ensures the deployment of The validation of the autonomous system in real cases of
added value services. It groups logical components for service provision made it possible to focus on the limitation of
negotiation with control entities as well as physical the SLM & M solution. In fact, the monitoring and correction
servers. operations make it possible to correct the QoS. However, the
correction time is very important. What influences is the
The strength of the NGN architecture lies in the ease of quality of user experience. For this purpose, the integration of
deployment of services without impacting the access networks technical choices to minimize time is not enough. This is
deployed by the operator. However, such an organization explained by the difficulty of the automatic correction
requires the intervention of several heterogeneous entities to operation and in real time. To this end, migration to a proactive
achieve the usual operations including user authentication and system becomes a necessity to ensure a stable level of service
service provision. This poses challenges for managing end-to- assurance at all times.
end quality of service.
B. The SLM and M Solution
The SLM&M solution aims to automate QoS management
procedures in NGN networks. An automation that has become
possible through the integration of business processes of the
eTOM framework [11][12]. The eTOM framework is a set of
business processes designed to model the usual operations in
telecommunications, in particular the provision, insurance and
billing of services.
The SLM&M solution has enabled the implementation of
an autonomous system for the monitoring and correction of
QoS impairments in NGN networks. This system is capable of
estimating customer satisfaction in real time based on the
Service Level Agreement (SLA) contract establishes when
subscribing the customer to the service provided [6]. If a
degradation is detected, the system proceeds to the correction
based on preconfigured scenarios, as the case encountered [13].
The autonomous system resulting from the SLM&M
solution consists of several modules (Fig. 1). The modules
include a set of eTOM business processes. Each process is
responsible for a specific activity such as collecting metrics,
loading fix configurations, and checking constraints defined in
SLA.
Fig. 1 illustrates the system architecture of the SLM & M
solution which consists essentially of two levels:
Monitoring Level: The first level includes business
processes that have a global view of the network. These
processes are responsible for detecting degradations and
identifying configurations to correct the identified case.
This level essentially includes three Services,
Orchestration and Resource modules. At this level the
communication between the modules is based on the
SOAP protocol. A justified and validated choice [9] to
facilitate the exchanges between the processes and to
minimize the time of execution.
Resource Level: The Resource Level has three modules.
These modules are responsible for collecting
performance indicators and implementing
configurations during the correction. Communication
between these modules is provided by the CORBA
protocol [14]. A choice that aims to ease trade [9]. Fig. 1. System Architecture of the SLM and M Solution.
593 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
III. PROACTIVE SYSTEM FOR INSURANCE OF SERVICES according to the domain and type of data processed. In our
The goal of the new approach is to migrate to a proactive context, the idea is to predict future values based on the old
system that can predict the state of the network. Once the values collected in regular time intervals. Also, the estimated
prediction is assured the second step is to run the SLM&M values must be done in a future time are sufficient to ensure the
system to proceed with the implementation of the necessary correction operation before the actual degradation of the
configurations to avoid the degradation. Migrating the SLM & service.
M solution to a proactive system is a difficult operation that For this purpose, time series [17] are the most appropriate
requires multiple tasks (Fig. 2). technique for our context. In fact, the time series make it
Fig. 2 outlines the steps required to implement the possible to model the values in time at regular intervals. Then
proactive approach. Steps that look like this: these values are modeled by a mathematical model in order to
be able to calculate values at times in the near future based on
Choice of indicators to use: The choice of indicators to the values recorded in the past times.
be estimated is a key step for the success of the
The implementation of a prediction model based on time
approach. Indeed, the prediction operation is a difficult
series requires the use of a suitable mathematical method.
operation that has a major blow in terms of time and
There are three methods in the literature for predicting values
resource.
based on time series, in particular:
Identification of the prediction model: There exists in
Moving Average [18]: An effective method for small
the literature a multitude of mathematical techniques
change values. IT is to model the series in a linear
and models for prediction in different disciplines. The
association with the moving averages of the process
model is the key to the success of the predictive
approach, given its effect on the accuracy of the Exponential Smoothing [19]: Exponential Smoothing is
predictions and thus the proactive solution. a highly valued tool for predicting and analyzing data
from time series observation. A technique used in
Integration of the model in the SLM & M solution:
industry, especially in financial markets, given the
After identifying the model it is necessary to interface
simplicity of the models included. Methods that are
the proactive system with the SLM & M solution. This
applicable to any discrete observation set. There are two
operation must ensure a transparent interface between
types of smoothing: (i) simple exponential smoothing,
the modules to ensure the proper functioning of the
(ii) double exponential smoothing.
system.
Box-Jenkins model [20]: Box-Jenkins provides an exact
Testing and Validation: The final step is the validation
methodology for identifying the time series model
of the evaluation of the approach in real cases of service
based on recorded observations. A methodology that is
provision.
based on several mathematical foundations to lead to a
A. Choice of Performance Indicators generally powerful prediction model.
Real-time monitoring is the collection of performance
indicators from many resources. These indicators depend
essentially on the outstanding service. For this purpose, the Selecting key
most fragile services are those of streaming type whose nature
of flows requires regularity over time. The indicators used in performance indicators
this type of service belong to two categories. (i) Static
indicators such as the codec used for each stream, the type of
video, the type of service (VoD, IPTV) as well as the capacity
of the server. (ii) Dynamic indicators such as jitter, delay,
percentage of lost packets, and actual throughput.
The multitude of indicators used will undoubtedly affect
the prediction time of degradation. For this, we have reduced
Prediction model
the performance indicators used for the video stream, since it is definition
the most sensitive to degradation. Indeed, in practical cases the
degradation affects the video stream before the audio. Only the
dynamic indicators will be estimated, since the static indicators
remain unchanged over time. Thus, the performance indicators
that will be processed in the model are (i) the jitter for the
video stream, (ii) the delay for the video stream; (iii) the
percentage of packets lost for the video stream. Implementation &
B. Choice of the Prediction Technique Validation
Several works have focused on the study and
implementation of prediction models in different domains Fig. 2. Description of the Methodology followed for the Migration to the
[15][16]. Work that offered a wide choice of models to adapt Proactive System.
594 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
gˆ t h j 0 1 g t _ j
the flow that changes for each session, which requires the use t 1 j
of a model capable of correcting itself automatically. Unlike
the ARMA method, exponential smoothing requires little (4)
computation, which will affect the resources less during the
Such as:
prediction operation. In addition, the method of smoothing has
demonstrated its effectiveness in various fields. So, our choice 1 1
was oriented towards exponential smoothing. N
625 10 (5)
C. Identification of the Prediction Model
The implementation of a simple exponential smoothing With: N- is the number of competing flows in the current
prediction model consists in adapting the variable α session.
(equation 1) according to the nature of the variation of the
TABLE I. COMPARISON BETWEEN MODELING APPROACHES OF TIMES
values of each indicator [19]. SERIES
xˆn ,t j 0 1 xn _ j
n 1 j
The costs of
Method Efficacy Adaptability
implementing
Very Costly in time and
The accuracy of the values estimated in the simple ARMA
Performance resource
Very difficult
exponential smoothing method depends on the adjustment of
the constant which takes values in the interval [0.1]. In the Exponential
Performance Fast Difficult
Smoothing
literature the study of the variation of the values makes it
possible to adjust the constant values. However, in our context Moving Average Average Fast Easy
the evaluation of the values of each indicator varies according
to the current session (Fig. 3). This forces the search for a self-
adaptable model according to the current situation.
Fig. 3 illustrates an example of observing the delay values
in three different sessions. Values that record large variations
in all three sessions. This difference does not allow the use of a
single constant for different sessions. Indeed, it is important to
adjust the value of the constant alpha so as to minimize the
error between the predicted value and the actual value (Fig. 4).
Fig. 4 illustrates the difference in the optimal value of the
constant for each flow in three different sessions. This shows
the difficulty of using the conventional method for
identification. For this purpose, we propose a model able to
justify the value of the constant according to the nature of the
current session for the two indicators delay and jitter. Indeed,
the experimental study showed that the value of the constant
depends on the number of flows in progress in the session
(Fig. 5).
In order to propose a model capable of self-justification, we
propose to establish a relation between α and the number of
flows noted 'N' using linear regression [21]. Before replacing
the constant by the formula in the prediction model; following
this method, we obtained the following prediction models:
1) Delay prediction model
dˆt h j 0 1 dt _ j
t 1 j
Fig. 3. Example of the Variation of the Delay Values in different Sessions.
595 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Access Agent
Estimator d Resource
d estim estimator Problem
Fig. 4. Adjustment of the Constant Alpha for Three Sessions. D Management
g estim estimator g Process
G
P estim estimator p
P
Resource
Resource Data collection Provisioning
Proceding Process Process
596 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
IV. EXPERIMENTS AND RESULTS Fig. 8 illustrates the quality of the video captured in the first
The experimentation phase aims at validating the prediction experimental case. For this purpose, the quality is very
approach in real cases of service provision in the NGN acceptable, given the absence of competing flows in the
network. In order to focus on the accuracy of the predicted network. Thus, the SLA contract of the customer is perfectly
values but also on the impact of such an approach on the respected by the service provider.
service assurance mechanism and the execution costs on the For this purpose, we note that the quality of video remains
resources. For this purpose, we propose the test bench acceptable despite the presence of competing flows in the
schematized in Fig. 7 that can emulate an NGN network and network. This is explained by the prediction of the indicators
the various modules of the proactive system. which made it possible to launch the configurations of the
The test bench consists of the following entities: corrections in order to avoid the effective degradation of the
quality of experience of the user BOB (Fig. 9).
Linux router in the boundaries that connect the core of
After validation of the proactive approach it is important to
the network to access networks. In addition, these
evaluate the impact of the prediction on resources. Fig. 10
routers also include the control entities deployed by the
illustrates the rate of CPU consumption in routers according to
OpenIMScore [22] solution as well as the Acess
the number of flows in the network after deployment of the
module of the monitoring solution.
prediction system. It can be noted that the prediction of the
Core-type linux router that includes other control indicators does not significantly affect the resources, since a
functions as well as some module of the platform difference of less than 7% is recorded in the most critical cases
solution. This router provides connectivity with the (160 flows). This is explained by the choice of exponential
application servers. smoothing for the implementation of the prediction model.
Management Server that includes the modules of the
solution belonging to the insurance level.
Server Streaming application server of type VoD.
The nominal flow of the experiment is carried out
according to two stages:
Case 1: At first, the BoB client whose SLA is of the
platinum type registers in the network before requesting
the VoD service.
Case 2: In a second step, several competing flows are
launched in the network via the IPREF [23] solution.
In both experimental cases, the proactive solution is
evaluated according to two essential criteria, namely the
accuracy of predictions and the cost of deploying resources.
The effect of the prediction of the indicators on the quality of Fig. 8. Screen Video Capture in the Second Test Case.
the experiment is also taken into consideration during the
experimental phase.
CPU Consumption (%)
80
70
60 without
predictio
50
n
40 with
predictio
30 n
20
10
0
#10 #40 #80 #100 #120 #160 #200
Fig. 7. Illustrates a Screenshot of the Video in the Second Experimental Fig. 9. CPU utilization by Routers According to the Number of Streams.
Case.
597 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
598 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
Abstract—The fourth-order finite difference Iterative The studies showed that the accuracies of the IADEDY and
Alternating Decomposition Explicit Method of Mitchell and the IADEMG are comparable to the IADEMF. Alias [5]
Fairweather (IADEMF4) sequential algorithm has demonstrated studied the parallel implementation of the IADEMF on
its ability to perform with high accuracy and efficiency for the distributed parallel computing using the parallel virtual
solution of a one-dimensional heat equation with Dirichlet machine. A fragmented numerical algorithm of the IADEMF
boundary conditions. This paper develops the parallelization of method was designed by Alias [6] in terms of the data-flow
the IADEMF4, by applying the Red-Black (RB) ordering graph where its parallel implementation using LuNA
technique. The proposed IADEMF4-RB is implemented on programming system was then executed. Sulaiman et al. [7, 8]
multiprocessor distributed memory architecture based on
proposed the half-sweep and the quarter-sweep IADEMF
Parallel Virtual Machine (PVM) environment with Linux
operating system. Numerical results show that the IADEMF4-
methods respectively, for the purpose of achieving better
RB accelerates the convergence rate and largely improves the convergence rate and faster execution time than the
serial time of the IADEMF4. In terms of parallel performance corresponding full-sweep method. Alias [9] implemented the
evaluations, the IADEMF4-RB significantly outperforms its Interpolation Conjugate gradient method to improve the
counterpart of the second-order (IADEMF2-RB), as well as the parallel performance of the IADEMF. Shariffudin et al. [10]
benchmarked fourth-order classical iterative RB methods, presented the parallel implementation of the IADEDY for
namely, the Gauss-Seidel (GS4-RB) and the Successive Over- solving a two-dimensional heat equation on a distributed
relaxation (SOR4-RB) methods. system of Geranium Cadcam cluster (GCC) using the Message
Passing Interface.
Keywords—Fourth-order method; finite difference; red-black
ordering; distributed memory architecture; parallel performance A recent study made by Mansor [11] involved the
evaluations development of a convergent and unconditionally stable
fourth-order IADEMF sequential algorithm (IADEMF4). The
I. INTRODUCTION proposed scheme is found to be capable of enhancing the
The heat equation is a mathematical model that describes accuracy of the original corresponding method of the second-
heat conduction processes of a physical system. Sahimi et al. order, that is, the IADEMF2. The IADEMF4 seems to be
[1] had proposed a finite difference scheme known as the more accurate, more efficient and has better rate of
Iterative Alternating Decomposition Explicit (IADE) method convergence than the benchmarked fourth-order classical
to approximate the solution of a one-dimensional heat iterative methods, namely, the Gauss-Seidel (GS4) and the
equation with Dirichlet boundary conditions. The IADE successive over-relaxation (SOR4) methods. However, the
scheme employs the fractional splitting of the Mitchell and IADEMF4 may be too slow to be implemented especially
Fairweather (MF) variant whose accuracy is of the order, when the problem involves larger linear systems of equations.
It is thus justified to consider parallel computing to speed up
O (t )2 (x)4 . The scheme, commonly abbreviated as the the execution time without compromising its accuracy. The
IADEMF, is developed by applying the second-order spatial algorithm has explicit features which add to its advantage,
accuracy to the heat equation. Due to the latter, in this paper, thus it can be fully utilized for parallelization.
the IADEMF will also be referred to as the IADEMF2. It is a This paper attempts to parallelize the IADEMF4, by
two-stage iterative procedure and has been proven to have applying the Red-Black (RB) ordering technique, for solving
merit in terms of convergence, stability and accuracy. It is large sparse linear systems that arise from the discretization of
generally found to be more accurate than the classical the one-dimensional heat equation with Dirichlet boundary
Alternating Group Explicit class of methods [2]. conditions. It aims to effectively implement the IADEMF4-
Several studies have later been developed based on the RB on parallel computers, with improved performance over its
IADE method. Sahimi et al. [3, 4] developed new second- serial counterpart. The high computational complexity of the
order IADE methods using different variants such as the IADEMF4-RB will be implemented on multiprocessor
D‟Yakonov (IADEDY) and the Mitchell-Griffith variant distributed memory architecture based on Parallel Virtual
(IADEMG). Each variant is of the order, O ( t )2 ( x )4 . Machine (PVM) environment with Linux operating system.
599 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
This paper is outlined as follows. Section II recalls the of boundary values and known u values at the previous time
formulation of the IADEMF4 scheme. Section III presents the level k .
development of the IADEMF4-RB parallel strategy. The
computational complexity of the RB methods considered in Au f
this paper is given in Section IV. Section V shows the
numerical experiment conducted in this study. The results and c d e u2 f2
u f
discussion on parallel performance of the methods under b c d e 3 3
consideration are discussed in Section VI. At the end of this a b c d e . .
paper is the conclusion.
. .
II. FORMULATION OF THE IADEMF4 (AN OVERVIEW) a b c d. e . .
In this section, the development of the IADEMF4
algorithm [11] is briefly reviewed. Consider the one-
a b c d um 2 fm2
f
dimensional heat equation (1) which models the flow of heat a b c ( m 2) x ( m 2) um 1 k 1 m 1
in a homogeneous unchanging medium of finite extent, in the (6)
absence of heat source. The entries in f are defined as
U U 2
f 2 b(u1k u1k 1 ) cu
ˆ 2k du3k eu4k
2
t x (1) f3 a(u1k u1k 1 ) bu2k cu
ˆ 3k du4k eu5k
subject to given initial and Dirichlet boundary conditions
fi auik 2 buik1 cu
ˆ ik duik1 euik 2 , i 4,5,..., m 3
U ( x, 0) f ( x), 0 x 1
ˆ mk 2 dumk 1 e(umk umk 1 )
f m 2 aumk 4 bumk 3 cu
U (0, t ) g (t ), 0 t T
ˆ mk 1 d (umk umk 1 )
f m 1 aumk 3 bumk 2 cu
U (1, t ) h(t ), 0 t T (7)
(2)
The IADEMF4 scheme secondly employs the fractional
Based on the finite difference approach, the time-space
splitting of the higher-order accuracy formula of the MF
domain is discretized by using a set of lines parallel to the t
variant [13],
axis given by xi ix , i 0,1, ... , m, m 1 and a set of lines
parallel to the x axis given by tk k t , k 0,1, ... , n, n 1 (rI G1 )u( p1/ 2) (rI gG2 )u( p) f (8)
. The grid spacings have uniform size, that is, x 1/(m 1)
and t T /(n 1) . At a grid-point P ( xi , tk ) in the solution (rI G2 )u( p1) (rI gG1 )u( p1/ 2) gf (9)
domain, the dependent variable U ( x, t ) which represents the
where G1 and G2 are two constituent matrices and r, I
non-dimensional temperature at time t and at position x, is
and p represent an acceleration parameter, an identity matrix
approximated by uik .
and the iteration index respectively. The value of g is defined
The IADEMF4 is developed by firstly executing the 6r
unconditionally stable fourth-order Crank-Nicolson as g , r 0 . The vectors u( p1) and u ( p1/ 2)
6
approximation (3) on the heat equation [12]. represent the approximate solution at the iteration level
1 k 1 k 1 1 ( p 1) and at some intermediate level ( p 1/ 2) ,
(ui ui ) ( x2 x4 )(uik 1 uik ) respectively.
t 2(x)2 12
(3)
After some algebraic manipulations for the equations in (8)
The discretization of (3) leads to the expression given in 1
(4), with the constants defined as in (5). and (9), the form, G1 G2 G1G2 u f is obtained,
6
auik21 buik11 cuik 1 duik11 euik21 auik 2 buik1 suggesting that matrix A in (6) can be decomposed into.
cu
ˆ ik duik1 euik 2 , i 2,3, ..., m 1 (4) 1
A G1 G2 G1G2
6 (10)
2 4 5 2 4 5
a , b , c , d , e , cˆ
24 3 4 3 24 4 (5) To retain the penta-diagonal structure of A , the matrices
G1 and G2 have to be in the form of lower and upper tri-
In matrix form, the approximation in (4) can be
represented by Au f (6), where A is a sparse penta- diagonal matrices respectively, Thus,
diagonal coefficient matrix, and the column vectors
u (u2 , u3 ,..., um2 , um1 )T contain the unknown values of u at
the time level k 1 and f ( f 2 , f3 ,..., f m2 , f m1 )T consists
600 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
6d li 1vˆi 1 6a 6b mˆ i 1uˆi 1
uˆi , mˆ i 1 , li ,
5 6 eˆi 1 6 eˆi
6(c 1) li uˆi mˆ i 1vˆi 1
eˆi 1
5 (12)
Since G1 and G2 are three banded matrices, then it is easy
to obtain the inverses of (rI G1 ) and (rI G2 ) . By
rearranging the equations in (8) and (9), the following Fig. 1. Computational Molecule of the IADEMF4 at the ( p 1/ 2) Iteration
expressions are obtained. Level.
u( p1/ 2) (rI G1 )1 (rI gG2 )u( p) (rI G1 )1 f (13)
u( p1) (rI G2 )1 (rI gG1 )u( p1/ 2) g (rI G2 )1 f (14)
Based on the above two equations, the computational
formulae at each of the half iteration levels can be derived as
given in (15) and (16).
1) At the (p+1/2) iteration level:
1
ui( p 1/2) ( Ei 1ui( p ) Wi 1ui(p1) Vi 1ui(p2) mˆ i 3ui(p21/2)
R
Fig. 2. Computational Molecule of the IADEMF4 at the ( p 1) Iteration
li 2ui(p11/2) fi ), i 2,3,..., m 2, m 1 Level.
(15)
601 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
computation of the unknown um( p11)i requires the values of for i 2, 4,.., m 2 (red grid-points) and i 3,5,.., m 1
(blackgrid-points)
um( p21)i and um( p31)i (Fig. 2). The unknown grid-points can only
be determined after the values of their two previous neighbors The purpose of including the relaxation factors y and
at their respective current iteration levels have been z in (18) and (19) is to accelerate the convergence rate of the
calculated. In other words, all values at the ( p 1 / 2)th level scheme.
cannot be calculated independently and simultaneously, so as The IADEMF4-RB ordering, on say, three processors, P 1,
values at the ( p 1)th level. These situations show that the P2 and P3, is illustrated in Fig. 3. P1 and P3 holds boundary
IADEMF4 is not inherently parallel. Thus, to handle this values at i 0 and i m 1 , respectively. The fourth-order
problem, this study resorts to undertake a domain methods require additional boundary values which are at
decomposition approach that firstly divides the physical positions i 1 (a grid-point in P1) and i m (a grid-point in
domain into a number of subdomains, each being assigned to a P3). As a strategy to obtain good load balancing, similar
processor; and secondly exchanges appropriate data across the numbers of alternate red (R) and black (B) grid-points are
boundaries of the subdomains. The Red-Black (RB) ordering assigned to each processor [18]. Depending on the color of the
is the domain decomposition strategy that is considered in this grid-point, the first two starting grid-points in a processor may
study. The approach focuses on minimizing the problem of be labelled as „stR‟ and followed by „stB‟, and the last two end
data dependencies and it is highly parallel. grid-points may be labelled as „enB‟ followed by „enR‟.
A. The IADEMF4-RB The following describes the implementation of the
The RB ordering has shown its competitiveness in terms of IADEMF4-RB based on Fig. 3. The algorithm is subjected to
speedup and efficiency, as has been proven in studies made the given initial and boundary conditions. Before the
by Evans [14] in solving the parallel SOR iterative methods; beginning of the execution, the unknowns, ui( p 1/2) , for
Brill et al. [15] in using the block GS-RB on the Hermite
collocation discretization of partial differential equations in i 2,3,.., m 1 , are given „guessed‟ values at the initial time.
two spatial dimensions; and Alias [5] in parallelizing the Then, the execution of the IADEMF4-RB algorithm is
IADEMF2. Darwis et al. [16] proved that the GS-RB performed in two phases:
algorithm is more accurate and converges faster than the GS The first phase involves the computations of only the red
algorithm. Yavneh [17] showed that the SOR-RB is more grid-points at the iteration levels ( p 1 / 2) and ( p 1) . This
efficient and smoother than the sequential SOR method for
solving two-dimensional Poisson equations. phase requires every processor to compute in parallel the red
unknowns by making use of the initialized „guessed‟ values.
This section parallelizes the IADEMF4 by using the RB Example, the computation of ust( p 1/2) in P2 requires „guessed‟
ordering technique. The algorithm used will be referred to as R
( p 1/2)
the IADEMF4-RB. uen value from P1 and ust( p 1/2) value from P2 itself, while
R B
( p 1)
The strategy to develop the IADEMF4-RB algorithm the computation of uen in P2 requires „guessed‟ ust( p 1)
begins by decomposing the domain into two different R R
( p 1)
independent subdomains, R and B . Each grid-point in value from P3 and uen
B
value from P2 itself.
the subdomains R and B is denoted red and black
respectively. If i is even, the grid-point is marked red, and if
i is odd, the grid-point is marked black. Assuming m is even,
then, the computational formulae for the IADEMF4-RB are:
y
ui( p 1/2) (1 y )ui( p ) ( Ei 1ui( p) Wi 1ui(p1) Vi 1ui(p2)
R
mˆ i 3ui(p21/2) li 2ui(p11/2) fi )
(18)
for i 2, 4,.., m 2 (red grid-points) and i 3,5,.., m 1
(blackgrid-points) Fig. 3. One-Dimensional IADEMF4-RB Ordering
602 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
After the computations of the red grid-points for the two IADEMF4 –RB: Slave’s Parallel Algorithm
iteration levels have been completed, adjacent processors begin
exchange their updated red values at the boundary grid-points slaves receive data from master: , ,
to prepare for the calculation of the black grid-points in the
second phase. Example, for
( p 1/2) determine initial conditions
Send updated uen : from P1 to P2, and from P2 to P3.
R
initialize guessed values
Send updated ust( p 1) : from P2 to P1, and from P3 to P2. end-for
R
while (time level < )
The second phase continues by computing simultaneously for and
the black unknowns at levels ( p 1/ 2) and ( p 1) , using the determine boundary conditions at ,
most recent red values computed in the first phase. For , and
example, the computation of ust( p 1/2) in P2 uses the updated red end-for
B
The updated black grid-points at the boundaries are then compute (refer to (7))
shared between adjacent processors. Example, end-for
set iteration
( p 1/2)
Send updated uen : from P1 to P2, and from P2 to P3 while (convergence conditions are not satisfied)
B
for
Send updated ust( p 1) : from P2 to P1, and from P3 to P2. compute
B
(refer to (18))
The two phases are repeated until convergence is reached. end-for
Due to the dependencies on the updated values between for
adjacent processors, the IADEMF4-RB algorithm involves
compute (refer to (19))
statements that take care of the communication between the
processors. An example of a procedure for sending and end-for
receiving messages between processors in a PVM send and receive updated red boundary
environment is as illustrated in Fig. 4. The IADEMF4-RB values between adjacent slave
processors (Fig. 4)
algorithm implemented by a slave processor can be described
as in Fig. 5. for
compute
if (left!=0) /* If there is a processor on the left*/ (refer to (18))
pvm_initsend( PvmDataDefault ); end-for
pvm_pkdouble( & [start], 1,1); for
pvm_send(left,50 ); compute (refer to (19))
end-if end-for
send and receive updated black
if (right!= 0) /* If there is a processor on the right*/ boundary values between adjacent
pvm_recv(right,50); slave processors (Fig. 4)
pvm_upkdouble(& [end+1],1, 1 ); test for convergence:
pvm_initsend( PvmDataDefault ); compute for
and
pvm_pkdouble(& [end], 1,1);
pvm_send(right,60 ); if max
end-if then
if (left!=0) /* If there is a processor on the left*/ add 1 to iteration (if necessary)
end-while
pvm_recv(left,60); end-while
pvm_upkdouble(& [start-1],1, 1 ); Determine numerical errors for and
end-if slave sends data analysis to master
pvm_exit;
end
Fig. 4. Communication Procedures for Sending and Receiving Messages
between Adjacent Processors. Fig. 5. IADEMF4-RB–Slave‟s Parallel Algorithm.
603 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
B. Parallel Algorithms for Benchmarking the details associated with message passing between
The IADEMF2, the GS4 and the SOR4 algorithms [9] can processors. From the memory perspective, the size of memory
also be parallelized using the RB ordering technique. They increases in proportion to the increasing number of processors.
will serve as the benchmarks for the parallel IADEMF4-RB. The parallel performances of the proposed algorithm was
The following are the schemes under consideration, assuming examined by solving a very large problem size on the
m is even. experiment in (23), where m varied from 70,000 to 700,000.
1) the IADEMF2-RB algorithms: This problem was taken from Saul‟yev (1964),
y U 2U
ui( p 1/2) (1 y )ui( p) (li 1ui(p11/2) si ui( p) 2 , 0 x 1
d t x (23)
wi ui(p1) fi )
(20) subject to the initial condition U ( x,0) 4 x(1 x),
for i 2, 4,..., m (red grid-points) and i 1,3,5,..., m 1 0 x 1 and the boundary conditions
(black grid-points) U (0, t ) U (1, t ) 0, t 0.
( p 1/2) ( p 1)
U ( x, t )
sum 1i gf m1i uˆm1i um 2 i )
3
k 1,(2) k 3
(21) (24)
for i 2, 4,..., m (red grid-points) and i 1,3,5,..., m 1 The other parameters considered for the experiment were
(black grid-points) = 0.5, t = 1.0204 x 10-12, t = 5.1020 x 10-11, and a stringent
2) the SOR4-RB algorithm (reduces to the GS4-RB tolerance value of 1015 . The initial and Dirichlet
algorithm when 1 : boundary conditions at i 0 and i m 1 were applied based
on the values given in the problem. For the fourth-order
ui( p 1) (1 )ui( p ) ( fi aui(p21) bui(p11) dui(p1) eui(p2) ) methods, the boundary values at positions i 1 and i m
c (22) were taken from the given exact solutions (24). The optimum
for i 2, 4,..., m 2 (red grid-points) and i 3,5,..., m 1 values for r and the relaxation factors ( x , y and ) were
(black grid-points) determined by experiments.
IV. COMPUTATIONAL COMPLEXITY VI. RESULTS AND DISCUSSION
The computational complexity of the RB algorithms of Table II compares the accuracy of the tested parallelized
interest is as given in Table I. It gives the number of parallel RB algorithms for a fixed problem size, m 700,000 . It is
arithmetic operations that is required to evaluate the obvious that the IADEMF4-RB outperforms the IADEMF2-
algorithms. RB in terms of rate of convergence. The average absolute
error, root mean square error and the maximum error of both
TABLE. I. PARALLEL ARITHMETIC OPERATIONS ( m PROBLEM SIZE, algorithms seem identical up to four decimal places, due to the
n NUMBER OF ITERATIONS, P NUMBER OF PROCESSORS) stringent tolerance value set in the experiment. The high
Number of Number of Total operation
computational complexity of the IADEMF4-RB is
Method compensated by the high accuracy it achieves at every
additions multiplications count
iteration and time level, causing its convergence to accelerate.
IADEMF4-RB 10(m-2)n / P 13(m-2)n / P 23(m-2)n / P
The SOR4-RB speeds up the convergence of the GS4-RB, but
IADEMF2-RB 6mn / P 9mn / P 15mn / P they are both relatively not reliable in terms of accuracy.
GS4-RB 4(m 2)n / P 5(m 2)n / P 9(m 2)n / P Table III displays the number of iterations ( n ), execution
time, speedup and efficiency of the IADEMF4-RB on using
SOR4-RB 5(m 2)n / P 7(m 2)n / P 12(m 2)n / P three different values of problem size, m. The execution time
refers to the amount of time required to complete a parallel
V. NUMERICAL EXPERIMENT program on a number of P processors from the moment the
The IADEMF4-RB was implemented and tested on execution starts till the moment the last processor finishes its
multiprocessor distributed memory architecture comprising of execution [19]. Speedup expresses how much faster the
twelve interconnected processors with Linux operating system parallel program executes relative to the sequential one.
using the PVM communication library. In distributed memory, Amdahl‟s law states that there exists a bound on the speedup
each processor has its own address space or local memory for a given problem with a fixed size [20], since some parts of
which is inaccessible to other processors. The processors the computations for solving a given problem are not
operate independently in parallel, and they share their data by parallelizable. Efficiency is a measure of the speedup achieved
means of some form of inter-processor communication via an per processor. It estimates how well the processors are utilized
inter-connection network. The programmer is responsible for during the execution of a parallel algorithm.
604 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
TABLE. II. PARALLEL RB ALGORITHMS – ERRORS AND NUMBER OF terms of execution time and speedup over serial computation.
ITERATIONS
Due to overheads, the overall efficiency for any m tends to
Root decrease as the number of processors increases. Overheads
Average Number have impacts on parallel performance. The two common types
Method mean Max.
absolute of
(m=700,000)
error
square error
iterations
of overheads are the communication time and the idle time.
error The communication time is the time spent on communication
IADEMF4-RB
and exchanging of data during the execution in all processors
1.5920e- 7.3054e- 1.9845e- and the idle time is the time when processors stay idle, waiting
r 0.8, y 1, z 1.1 09 09 07
288
for busy processors to send messages. Idling may be due to
load imbalances amongst processors, or a bottleneck at the
IADEMF2-RB 1.5920e- 7.3054e- 1.9845e- master processor when it has to interact with other worker
r 0.8, y 1, z 1.1 09 09 07
450
processors [21].
For every number of processor ran in the experiment, the
SOR4-RB 1.6150e- 9.6395e- 2.7422e-
( 1.06 ) 09 09 06
738 execution time for a problem size of 70,000 is comparatively
smaller than a problem ten times its size. This is expected
since fewer grid-points involve less mathematical operations
1.6150e- 9.6395e- 2.7422e-
GS4-RB
09 09 06
794 and data sharing. The table, however, shows an improvement
in convergence rate, speedup and efficiency as the size
= 0.5, x = 2.60 x 10-6, t = 1.02 x 10-12, t = 5.10 x 10-11, = 1 x 10-15 increases to 700, 000 . The smaller size with higher number of
iterations (n) seems to be less efficient due to the additional
TABLE. III. IADEMF4-RB – PERFORMANCES USING SEVERAL VALUES OF m overhead imposed by having communications routed through
the PVM daemon.
Execution
m x P time (s)
Speedup Efficiency Fig. 6 shows that the execution time taken by every tested
1 4.869491 1 1
algorithm (listed in Table II) decreases with increasing P .
However, the IADEMF4-RB executes in the least amount of
2 2.507665 1.941843 0.970921 time for every P . Despite the IADEMF4‟s greater
4 1.518787 3.206171 0.801542 computational complexity, its parallelization using the RB
70,000
1.43 x 10 -5
6 1.261464 3.860190 0.643365
technique and the use of relaxation parameters have enabled it
n 359 to execute in a shorter time on one and more processors in
8 1.102297 4.417585 0.552198 comparison to its counterpart of second-order.
10 1.039360 4.685086 0.468508
Fig. 7 shows that every tested algorithm has a speedup of
12 1.008263 4.829584 0.402465 less than P , which implies that the parallel code is bounded by
1 20.039541 1 1 the sequential code (Amdahl‟s law). The parallel code runs
slower due to overheads that outweigh the benefits of parallel
2 10.062964 1.991415 0.995707
computation. Amongst the four algorithms, the IADEMF4-RB
4 5.300258 3.780842 0.945210 proves to continue giving the best speedup as P increases. At
385,000
2.60 x 10-6 6 3.828272 5.234617 0.872436 P 12 , the speedup of the IADEMF4-RB is almost 14%
n 312 closer to the linear speedup. As for the IADEMF2-RB, the
8 2.993626 6.694069 0.836758
SOR4-RB and the GS4-RB, there is an 18, 24 and 28 percent
10 2.447466 8.187873 0.818787 difference, respectively, between the method‟s speedup and
12 2.101530 9.535691 0.794640 the linear speedup.
1 35.682042 1 1
60
2 17.896741 1.993773 0.996886
50
4 8.962541 3.981241 0.995310 IADEMF2-RB
Execution time (s)
700,000
1.43 x 10-6 6 6.202509 5.752840 0.958806 40
n 288 IADEMF4-RB
8 4.900683 7.281034 0.910129
30 GS4-RB
10 3.992991 8.936168 0.893616
12 3.456841 10.32215 0.860179 20 SOR4-RB
605 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
14 1.2
12 1.0
10
0.8
Efficiency
Speedup
8
0.6 IADEMF2-RB
6
IADEMF2-RB IADEMF4-RB
0.4 GS4-RB
4 IADEMF4-RB
GS4-RB SOR4-RB
2 SOR4-RB 0.2 Optimum Efficiency
Linear Speedup
0 0.0
1 2 4 6 8 10 12 1 2 4 6 8 10 12
Number of processors Number of processors
Fig. 7. Speedup Versus Number of Processors. Fig. 8. Efficiency Versus Number of Processors.
Temporal performance
2.5E-01
speedup performance by the IADEMF4-RB (Fig. 7), however,
makes it the most efficient algorithm amongst the tested 2.0E-01
algorithms. With the number of processors equals to 12, the
1.5E-01
IADEMF4-RB achieves a speedup of 10.32 that equates to a
higher efficiency of about 0.86 (Table III). IADEMF2-RB
1.0E-01
IADEMF4-RB
Temporal performance is a metric which is inversely GS4-RB
5.0E-02
proportional to the execution time. If there are several parallel SOR4-RB
algorithms solving the same problem with the same problem 0.0E+00
size implemented on the same number of processors, then the 1 2 4 6 8 10 12
algorithm with the largest value for temporal performance will Number of processors
be considered as the best algorithm that can perform in the
least amount of execution time. Fig. 9 shows that the Fig. 9. Temporal Performance Versus Number of Processors.
IADEMF4-RB has proven itself as the algorithm with the best
temporal performance amongst all the methods considered for TABLE. IV. SUMMARY OF THE GRANULARITY RESULTS FOR THE TESTED
comparison. RB METHODS
606 | P a g e
www.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 10, No. 9, 2019
The proposed IADEMF4-RB parallel algorithm [7] J. Sulaiman, M. K. Hasan, and M. Othman, “The half-sweep Iterative
significantly outperforms its counterparts of the second-order, Alternating Decomposition Explicit Method (HSIADE) for diffusion
equations,” Lecture Notes on Computer Science, vol. 3314, Berlin-
as well as the benchmarked fourth-order classical methods. Heidelberg, pp. 57-63, 2004.
This is with regards to accuracy, convergence rate and parallel [8] J. Sulaiman, M. K. Hasan, and M. Othman, “Quarter-sweep Iterative
measures such as execution time, speedup, efficiency, Alternating Decomposition Explicit algorithm applied to diffusion
temporal performance and granularity. Despite its higher equations,” International Journal of Computer Mathematics, vol.
computational complexity, its increasing number of correct 81(12), pp. 1559-1565, 2004.
[9] N. Alias, M. S. Sahimi, and A. R. Abdullah, “Parallel strategies for the
digits at each iteration yields faster rate of convergence with Iterative Alternating Decomposition Explicit Interpolation-conjugate
higher level of accuracy for a large size matrix. The relatively Gradient method in solving heat conductor equation on a distributed
coarse granularity delivered by the RB parallel parallel computer systems,” Proceedings Third International Conference
implementation indicates the feasibility of parallelizing the Numerical Analysis Eng., pp. 31-38, 2003.
[10] R. H. Shariffudin and S. U. Ewedafe, “Parallel domain decomposition
proposed IADEMF4. for 1-D active thermal control problem with PVM,” International
The efficient performance in parallel gives benefits, Journal of Advanced Computer Science and Applications, vol. 6, No.
10, 2015.
especially in solving problems involving larger sparse linear [11] N. A. Mansor, A. K. Zulkifle, N. Alias, M. K. Hasan, and M. J. N.
systems of equations that usually consumes huge amount of Boyce, “The higher accuracy fourth-order IADE algorithm,” Journal of
serial time. Future work is to consider applying the Applied Mathematics, vol. 2013 Article ID 236548,
IADEMF4-RB in time-dependent PDEs that require higher- http://dx.doi.org/10.1155/2013/236548, 2013.
[12] G. D. Smith, “Numerical solution of partial differential equations: Finite
order accuracy with significant speedup and efficiency. difference methods,” second ed., Oxford University Press, 1978.
Another possibility is to apply the proposed parallel method [13] A. R. Mitchell and G. Fairweather, “Improved forms of the alternating
onto shared or hybrid memory architectures to reduce the direction methods of Douglas,Peaceman,and Rachford for solving
problem of communication issues. parabolic and elliptic equations,” Numerische Mathematik, vol. 6 (1),
pp. 285–292, 1964.
REFERENCES [14] D. J. Evans, “Parallel S.O.R iterative methods,” Parallel Computing, vol.
[1] M. S. Sahimi, A. Ahmad, and A. A. Bakar, “The Iterative Alternating 1, pp. 3-18, 1984.
Decomposition Explicit (IADE) method to solve the heat conduction [15] S. H. Brill and G. F. Pinder, “Parallel implementation of the Bi-
equation,” International Journal of Computer Mathematics, vol. 47, pp. CGSTAB method with Block Red-Black Gauss-Seidel preconditioner
219-229, 1993. applied to the Hermite Collocation discretization of partial differential
[2] D. J. Evans and M. S. Sahimi, “The Alternating Group Explicit Iterative equations,” Parallel Computing, vol. 28:3, pp. 399-414, 2002.
Method to solve parabolic and hyperbolic partial differential equations,” [16] R. Darwis, N. Alias, N. Yaacob, M. Othman, N. Abdullah, and T. Y.
Ann. Rev. of Num. Fluid Mechanics and Heat Transfer, vol. 2, pp. 283- Ying, “Temperature behavior visualization on rubber material involving
389, 1989. phase change simulation,” Journal of Fundamental Sciences, vol. 5, pp.
[3] M. S. Sahimi, E. Sundararajan, M. Subramaniam, and N. A. A. Hamid, 55-62, 2009.
“The D‟Yakonov fully explicit variant of the iterative decomposition [17] I. R. Yavneh, “On Red-Black SOR smoothing in multigrid,” SIAM J.
method,” Comp. Math. , vol. 42, pp. 1485-1496, 2001. Sci. Comput. , vol. 17(1), pp. 180-192, 1995.
[4] M. S. Sahimi, N. A. Mansor, N. M. Nor, N. M. Nusi, and N. Alias, “A [18] B. Körfgen and I. Gutheil, “Parallel linear algebra methods,
high accuracy variant of the Iterative Alternating Decomposition computational nanoscience: do it yourself!,” John von Neumann
Explicit method for solving the heat equation,” Int. J. Simulation and Institute for Computing. Jülich, NIC Series, vol. 31, pp. 507-522, 2006.
Process Modelling, vol. 2, Nos. 1/2, pp. 77-86, 2006. [19] J. Kwiatkowski, “Evaluation of parallel programs by measurement of its
[5] N. Alias, “Development and implementation of parallel algorithms in granularity,” Proceeding PPAM ‟01 International Conference on Parallel
the IADE and AGE class of methods to solve parabolic equations on a Processing and Applied Mathematics–Revised Papers, Springer-Verlag,
distributed parallel computer systems,” PhD Thesis, Universiti London, 2002.
Kebangsaan Malaysia (2003). [20] G. M. Amdahl, “Validity of the single-processor approach to achieving
[6] N. Alias and S. Kireev, “Fragmentation of IADE method using large scale computing capabilities,” AFIPS Conference Proceedings,
LuNA system,” Malyshkin V. (eds) Parallel Computing vol. 30. AFIPS Press, Reston, Va., pp. 483-485, 1967.
Technologies, Lecture Notes in Computer Science, vol. 10421. [21] J. Lemeire, “Leaning causal models of multivariate systems and the
Springer, Cham, 2017. value of it for the performance modeling of computer programs,” PhD
Thesis, Vrije Univesiteit, Brussel, Brussels University Press, 2007.
607 | P a g e
www.ijacsa.thesai.org