Beruflich Dokumente
Kultur Dokumente
Abstract—Future buildings will offer new convenience, com- time complicate the process of managing SBs. In particular,
fort, and efficiency possibilities to their residents. Changes will SBs and their inhabitants are expected to create large volumes
occur to the way people live as technology involves into people’s of streaming data. ML, sampling, compression, learning, and
lives and information processing is fully integrated into their
daily living activities and objects. The future expectation of smart filtering technologies are becoming more significant to manage
buildings includes making the residents’ experience as easy and the stream of big data of individuals.
comfortable as possible. The massive streaming data generated many other types of SBs
and captured by smart building appliances and devices contains In 1981, the term Intelligent Buildings (IBs) was initially
valuable information that needs to be mined to facilitate timely coined by United Technology Building Systems (UTBS) Cor-
actions and better decision making. Machine learning and big
data analytics will undoubtedly play a critical role to enable poration in the U.S. In July 1983, IBs became a reality with the
the delivery of such smart services. In this paper, we survey opening of the City Place Building in Hartford, Connecticut
the area of smart building with a special focus on the role of [1]. Today, the number of SBs is growing at an unprecedented
techniques from machine learning and big data analytics. This rate including smart office, smart hospitality, smart educational
survey also reviews the current trends and challenges faced in facilities etc. [2]. An SB is recognized as an integrated system
the development of smart building services.
that takes advantage of a range of computational and commu-
Index Terms— Smart Buildings, Smart Homes, Internet of nications infrastructure and techniques [3]. Examples of SB
Things (IoT), Big Data Analytics, Machine learning (ML).
services include smart thermostats that allow the temperature
to be controlled based on the time of the day/year and the
I. I NTRODUCTION users’ preferences with minimal or no manual configuration.
Although the term “smart building” (SB) may bring a Using data analytics to “learn” the users’ preferences before
thought of a fictional smart space from science-fiction movies, taking the appropriate actions is probably the most important
but the reality is that SBs exist today, and their number is enabling technology for IBs [4]. Lately, smart coffee machines
getting increased. With recent advances in machine learning appeared in the market with the capability to make coffee
(ML), big data analytics, sensor technologies and the Internet automatically, according to users’ preferences and schedules.
of Things (IoT), regular buildings can be cost-effectively Fridges can offer allocated programming interfaces for their
transformed into SBs with bare minimum infrastructural mod- control [5]. IBs aim to provide their users with safe, energy
ifications. There are smart office, smart library, smart home, efficient, environment-friendly, and convenient services.
smart health care facilities, smart hospital and many other In order to maximize comfort, minimize cost, and adapt to
types of SBs that can provide automated services that can the needs of their inhabitants, SBs must rely on sophisticated
provide many value-added services (such as reduction of tools to learn, predict, and make intelligent decisions. SB
wasted energy) and also help to ensure the comfort, health, algorithms cover a range of technologies, including prediction,
and safety of the occupants. decision-making, robotics, smart materials, wireless sensor
The hyperconnectivity that will be brought about by the networks, multimedia, mobile computing, and cloud com-
emergence of IoT will increase the promise of SB since puting. With these technologies, buildings can cognitively
now all the basic building amenities and commodities ranging manage many SB services such as security, privacy, energy
from your house electronics to your plant vases will be efficiency, lighting, maintenance, elderly care, and multimedia
interconnected. But this hyperconnectivity will at the same entertainment.
The massive volume of sensory data collected from sensors
B. Qolomany, A. Al-Fuqaha, A. Gupta, S. Alwajidi, and
A. Fong are with the Department of Computer Science,
and appliances must be analyzed by algorithms, transformed
Western Michigan University, Kalamazoo, MI 49008 USA (e- into information, and minted to extract knowledge so that
mail: basheer.qolomany@wmich.edu; ala.al-fuqaha@wmich.edu; machines can have a better understanding of humans than
ajay.gupta@wmich.edu; safaakhalilmu.alwajidi@wmich.edu;
alvis.fong@wmich.edu). their environment. Furthermore, and most importantly, such
D. Benhaddou is with Engineering Technology Department, University of knowledge can lead to new products and services that can
Houston, Houston, Texas 77204 USA (e-mail: dbenhadd@central.uh.edu). dramatically transform our lives. For example, readings from
J. Qadir is with Information Technology University, Lahore, Pakistan (e-
mail: junaid.qadir@itu.edu.pk). smart meters can be used to better predict and balancing the
usage of power. Monitoring and processing sensory data from
wearable sensors attached to patients can produce new remote
healthcare services.
The main philosophy behind ML is to create the analytical
models automatically in order to permit the algorithms to
learn continuously from available data. The application of
ML techniques increased over the last two decades due to
the availability of massive amounts of complex data and
the increased usability of current ML tools. Today, ML is
already widely applied in different applications including
recommendation systems offered by online services (e.g.,
Amazon, Netflix) and automatic credit rating services used by
banks. Alphabet’s Nest thermostat utilizes ML to “learn” the
temperature preferences of its users and adapt to their work
Fig. 2: Hype Cycle for the Connected Home, 2018 [7].
schedule to minimize the energy use. Other widely publicized
examples of ML applications include Google’s self-driving
car, sentiment analysis of Amazon and Twitter data, fraud Gartner’s 2018 Hype Cycle expectation for deep learning, ML,
detection, and Facebook’s facial-recognition technology that connected homes, and smart workspace.
is used to tag the suggested person on images uploaded by According to report by Research and Markets [10][11], the
users. global IoT SB market will reach approximately $51.44B USD
globally by 2023. The report also forecast that 33% of IoT SB
market will be supplied by artificial intelligent technologies
A. SB Trends and Market Impact
by 2023, and automation systems of SB will grow at 48.3%
In this section, we look at the statistics related to SBs, to CAGR from 2018–2023. Frost & Sullivan also predict that
allow us to understand the current trends and motivations in by 2025, the growth of connected home living will reach 3.7
industry marketplaces and academic researches toward SBs. billion smartphones, 700 million tablets, 520 million wearable
According to the report by MarketsandMarkets [6], The SB health-related devices and 410 million smart appliances in the
market is estimated to grow from 7.42 billion dollars in 2017 to connected person world.
31.74 billion dollars by 2022, at a Compound Annual Growth
Rate (CAGR) of 33.7% from 2017 to 2022. In yet another B. Related Survey Papers
report Zion Market Research [7], 2016 and it is expected to
reach 61,900 million dollars by 2024. It is expected to exhibit a Although many of survey papers focused on SBs have been
CAGR of more than 34% between 2017 and 2024. The market published, none of them is focused on the role of data analytics
is primarily driven by government initiatives globally for SB and ML in the context of SBs. We describe the relevant survey
projects and the increasing market for integrated security and papers next and will compare these survey papers to our paper
safety systems as well as energy efficient building systems. in Table II.
Figure 1 shows the Statista [8] forecast market size of the • Chan et al. in 2008 provided an overview of smart
global smart home market from 2016 to 2022 (in billion U.S. home research [12]. It also discusses assistive robots, and
dollars). wearable devices. The article reviews smart home projects
arranged by country and continent.
• Alam et al. [13] provided details about sensors, de-
vices, algorithms, and communication protocols utilized
in smart homes. The paper reviews smart home works
according to their desired services and research goals;
namely, security, comfort, and healthcare.
• Lobaccaro et al. [14] presented the concept of smart home
and smart grid technologies and discuss some challenges,
benefits and future trends of smart home technologies.
• Pan et al. [15] reviewed the works on efficient energy
consumption in SBs using microgrids. The survey inves-
tigates research topics and the recent advancements in
SBs and the vision of microgrids.
• A few survey papers have reviewed works on facilitating
Fig. 1: Forecast market size of the global smart home market independent living of the elderly people in smart homes.
from 2016 to 2022 [8]. Ni et al. [16] conducted a survey on the features of
sensing infrastructure and activities that can assist the
According to the Gartner report [9], it is expected that the independent living of the elderly in smart homes. A
number of smart connected homes grows to 700 million homes survey on ambient assisted living technologies for elderly
by 2020, supplied by mass consumer adoption and an increase people has been presented Rashidi and Mihailidis [17].
in the number of devices and apps available. Figure 2 shows Peetoom et al. [18] focused on monitoring technologies
to recognize life activities in-home such as fall detection TABLE I: LIST OF IMPORTANT ACRONYMS USED
and changes in health status. Salih et al. [19] presented a AAL Ambient Assisted Living
review of ambient intelligence assisted healthcare mon- ANNs Artificial Neural Networks
AODE One-Dependence Estimators
itoring services and described the various application, APAC Asia and Pacific
communication, and wireless sensor network technolo- AR Accuracy Rate
BBN Bayesian Belief Network
gies that have been employed in the existing research BT-LE Bluetooth Low Energy
literature. CAGR Compound Annual Growth Rate
CAN Controller Area Network
• A number of papers have focused IoT: (a) Perera et al. CART Classification and Regression Tree
[20] discussed IoT applications from the perspective of CEA Consumer Electronics Association
CEP Complex Event Processing
context-awareness and self-learning; (b) Tsai et al. [21] CHAID Automatic Interaction Detection
surveyed the applications of data mining technologies in CNN Convolutional Neural Network
CNN Convolutional Neural Networks
IoT; and (c) Mahdavinejad et al. [22] reviewed some ML CRF Conditional Random Field
methods that can be applied to IoT data analytics. DBM Deep Boltzmann Machine
DBN Deep Belief Networks
DIY Do-It-Yourself
ECG Electrocardiography
C. Contributions and Organization of This Paper EEG Electroencephalography
EM Expectation Maximization
To the best of our knowledge, this is the first survey that EMG Electromyography
covers SBs jointly from the perspectives of application, data EMSs Energy Management Systems
analytics, and ML. The main contributions of our paper are: EOG Electrooculography
ET-KNN Evidence Theoretic Knearest Neighbors
• Exploration of the potential of ML-based context-aware FLS Fire and Life Safety
GBM Gradient Boosting Machines
systems to provide SB services; GBRT Gradient Boosted Regression Trees
• Identification of research challenges and directions for GSR Galvanic Skin Response
GUI Graphical User Interface
SBs and how ML models can help in resolving such HDFS Hadoop Distributed File System
challenges; HMMs Hidden Markov Models
HVAC Heating, Ventilation, and Air Conditioning
• Identification of SB applications including comfort, se- IBs Intelligent Buildings
curity, energy efficiency, and convenience and the role IBT Intelligent Building Technology
ICA Independent Component Analysis
of ML in such applications. Our research can provide an ICT information and communication technologies
impetus to ML researchers to investigate new exciting ID3 Iterative Dichotomiser 3
IoT Internet of Things
ML-based SB services. ISM Bands Industrial Scientific Medical Bands
The rest of the paper is organized as follows: Section II KNX Konnex
LANs Local Area Networks
introduces the concept of SBs and its underlying architecture. LCR Lighting Control and Reduction
Section III introduces the various components of the SB LDA Linear Discriminant Analysis
LOESS Locally Estimated Scatterplot Smoothing
ecosystem and its underlying architecture. Section V presents LoT Lab of Things
context recognition and activity modeling and the role of MARS Multivariate Adaptive Regression Splines
M-Bus Meter-Bus
ML in SBs. Section VI highlights research and development MISs Management Information Systems
challenges and provides a future perspective of SB projects. ML Machine Learning
NFC Near Field Communication
Finally, Section VII presents a summary of lessons learned NLP Natural Language Processing
and concludes the paper. OA Office Automation
OLSR Ordinary Least Squares Regression
For the convenience of the readers, we have enlisted the ORE Oracle R Enterprise
important acronyms used in Table I. OSX Oracle Stream Explorer
PCA Principal Component Analysis
PCR Principal Component Regression
II. SMART B UILDINGS : C ONCEPT AND ARCHITECTURE PLC Powerline Communication
PSNR Peak-Signal-to-Noise Ratio
In 1984, The New York Times published an article that RBFN Radial Basis Function Network
RBM Restricted Boltzmann Machine
described that real estate developers are creating “a new RDDs Resilient Distributed Datasets
generation of buildings that almost think for themselves called RFID Radio Frequency Identification
RNN Recurrent Neural Networks
intelligent buildings.” Such an intelligent building (IB) was SB Smart Building
defined as “a marriage of two technologies old-fashioned SVMs Support Vector Machines
TMSs Temperature Monitoring Systems
building management and telecommunications.” [23]. Since UTBS United Technology Building Systems
then, many definitions of SBs have been suggested. This is
due to the fact that the life-cycle of building planning, design,
implementation, and operation involves different industry play-
ers that have different roles. In addition, the rapid changes in of SBs [1]. The Institute for Building Efficiency [24] focuses
technology are affecting this definition. For instance, the ad- on the operation of buildings to provide efficient healthy and
vent of IoT and smart city concepts is impacting the definition comfortable environment [25]. IBM [26] focuses also on the
of SB. Therefore, it is hard to compose a unique view of IBs operation of SBs to provide integrated physical and digital
with a single definition that is accepted worldwide. However, infrastructures that provide reliable, sustainable, and cost-
it is vital to have a good understanding of the main standard effective occupancy services. According to the European Com-
bodies and companies involved in shaping the development mission’s Information Society [27], SBs means buildings that
TABLE II: COMPARISON OF RELEVANT SURVEY PAPERS
Cite Purpose Limitations
Chan et al. [12] Review SH projects arranged by country and continent as Does not focus on the role of ML and big data analytics, it
well as the associated technologies for monitoring systems does not review and categorize the papers according to the
and assistive robotics applications of SH
Alam et al. [13] Reviews SH projects according to research objectives and Does not focus on the role of ML and big data analytics for
services; namely, comfort, healthcare, and security. SB.
Lobaccaro et al. review of existing software, hardware, and communications Does not focus on the role of ML and big data analytics.
[14] control systems for SH and smart grid It also does not focus on reviewing and categorizing papers
according to the applications of SH.
Pan et al. [15] Review the research topics on the energy efficiency and the The focus of the paper is not the ML and big data analytics
vision of microgrids in SBs. for SB services. It also does not cover other applications of
SB rather than energy efficiency.
Ni et al. [16] propose a classification of activities considered in SH for Does not cover all the services in SH. It also does not
older peoples independent living, they also classify sensors categorize the research according to different ML model
and data processing techniques in SH. styles.
Rashidi and Mi- Review AAL technologies, tools, and techniques The paper focuses only on AAL in healthcare, and does not
hailidis [17] cover the other applications in SH or SB; in addition, there
is no classifying of the researches according to ML model
styles
Peetoom et al. Review the works on monitoring technologies that detect Does not focus on the role of ML in SB.
[18] ADL or significant events in SH.
Salih et al. [19] Review the works on ambient intelligence assisted healthcare The paper also does not show the challenges and the future
monitoring focuses only on AAL in healthcare, and does not research directions in the field.
cover the other applications in SH or SB.
Perera et al. [20] Review the works in context awareness from an IoT perspec- Does not focus specifically to the SB domain and its appli-
tive cation services.
Tsai et al. [21] Review the research works of data mining technologies for Does not focus specifically to SB applications.
IoT applications.
Mahdavinejad et Review some ML methods applied to IoT data by studying Does not focus on SB and its applications as a use case.
al. [22] smart cities as a use case scenario.
life and seasons change automatically. Just use it for a week of your home.
and it programs itself. It learns about the level of temperatures Blufitbottle [49] this bottle records the drinking habits of the
that the occupants prefer and creates a context-aware person- users and sends them notifications about the time and amount
alized schedule. The smart thermostat turns to an energy- of the water that they are supposed to drink to keep them
efficient mode automatically when the residents leave the healthy and hydrated. The app collects data about users such
building. It could start warming up the area when it senses as their weight and age, plus other factors such as the current
activity, such as an occupant’s returning back home from work. levels of temperature and humidity to estimate the amount of
The Nest Thermostat is controllable via a smartphone and an the needed water to keep them hydrated. When the user falls
installed app. If you are away for a while, this device has also behind with hydration, an alert sounds, as well as a simple
a capability to sustain a particular temperature in your house. glance from the LEDs, will indicate when it’s time for the
Lockitron [46] is a door lock that can control the door next drink.
remotely over the Internet to open and close it by phone. Canary [50] is an all-in-one home security system that
Lockitron app can be installed and used by any iOS or Android comprises a set of sensors such as temperature, air quality,
smartphone. Homeowners can directly grant family and friends sound, motion vibration, in addition, an HD video camera in
the access to open a given door by providing authorization over one unit. The system utilizes ML algorithms to let the users
the Internet. Lockitron can also utilize Bluetooth low energy know what is happening at home and take action by sending
technology, which means that it will keep controlling even notifications to your phone if something happens. Those ML
in the event of Internet or power outages. Lockitron can also models learn over time and send the users smarter notifications
connect to the Internet with Bridge, through which occupants as it detects motion. So that, the longer you have the system,
can control the bolt anywhere in the world. the more effective it becomes. Canary is able to decrease the
The SmartThings [47] SB automation system comprised rate of false alarms by learning the user behavior and the
of a communications smart hub, that supports various smart ambient noise level and the home temperature patterns.
appliances and devices; the smart hub supports various tech- Amazon Echo [51] is a small cylinder enable the users to
nologies and protocols such as ZigBee, Z-Wave, as well as IP- control anything in the home via the voice. Amazon Echo has
accessible devices and lets you control appliances using Wi-Fi a powerful voice recognition capability, the user does not have
and Bluetooth connectivity. SmartThings provides kits that to worry about the complexities of their voice. Amazon Echo
include smart plugs, in addition, the basic sensors that can be is connectable via Wi-Fi or Bluetooth, the users can send voice
utilized to measure temperature, as well as to detect presence, commands to control the speakers as well as other compatible
motion, orientation, and vibration. SmartThings also includes devices such as Belkin’s WeMo and Philips Hue. It can also
an open platform that enables smart device vendors and third- use Amazon cloud Lambda service to send commands. To send
party software to provide hardware and software that can be any command It requires to include the name of the program,
utilized alongside the platform. for instance, “Alexa, turn on TV”. It also includes a network
Philips Hue [48] is a combination of LED lighting with mo- to distant servers, which slows down the response time.
bile technology. An accompanying mobile app that allows you Honeywell Total Connect Remote Services [52] this device
to control lighting systems and changing color sets depending merges personal smart home automation with security mon-
on your mood utilizing Wi-Fi technology. The new Philips Hue itoring task. It enables the occupants to control and monitor
bridge supports the required authentication to enable Apple everything in the home from lighting and window shades sys-
HomeKit technology to control and enable your Philips Hue to tems to security cameras and smoke alarms. the user can utilize
connect to other HomeKit enabled accessories and take control a smartphone app or desktop-mounted hardware console for
TABLE IV: COMPARISON AMONG VARIOUS SMART CONTROL DEVICES IN SB
Technology Platform used Pros Cons
Affordable hardware options. Can expand
Android, iOS, No color bulbs, no dimmer switches. Ex-
WeMo Wi-Fi using SmartThings hub. Can expand us-
Window phone perienced some latency issues.
ing IFTTT.
More expensive than other smart ther-
iOS, or Android, Easy to program. It learns the user daily
Nest Wi-Fi, Zigbee, mostats. The Nest might not be for you
Mac OS, Win- routine, it could set itself up for the user
Thermostat Thread if you are a stickler for temperatures that
dows lifestyle after the first week of use.
are “just right”.
Affordable, easy to install, quiet opera-
Wi-Fi bridge costs extra. Does not work
Lockitron Bluetooth Android, iOS tion, Offers proximity locking and un-
with other devices.
locking.
Wi-Fi, Zigbee, Affordable. Easy to install. Quiet oper- Compatibility issues with other devices,
Android, iOS,
SmartThings Z-Wave, Blue- ation. Offers proximity locking and un- workarounds for non-natively supported
Window phone
tooth locking. devices will be difficult for some.
Modest list of 3rd party integrations.
Least user-friendly app. Complex to con-
Offers own proprietary hardware. Ac-
Philips Hue Wi-Fi Android, iOS figure simple tasks. Missing some “key”
tive web community for help. Deep cus-
integrations.
tomization for power users.
iOS and Android Easy to set up and use. Excellent light The Philips HUE kit, has a fairly high
Blufitbottle Bluetooth
device quality cost.
iOS or Android
Canary Wi-Fi Simple and easy to set up. Expensive.
device
Fire OS, High-quality voice recognition. Integra-
Amazon Wi-Fi, The system only detects the intruder once
Android, and tions with all of the “key players”. Works
Echo Bluetooth they are inside premises.
iOS for all users, regardless of phone brand.
Priced higher than other smart ther-
Honeywell Wi-fi, Z-Wave iOS and Android Easy installation.
mostats with similar functionality.
controlling and monitoring. It can provide real-time alerts, home gateway can also collect and store data for a specific
GPS vehicle and asset tracking, video viewing, and mobile time period. Typically, these gateways can connect to the cloud
control. The system only supports Z-Wave devices, it needs services and perform data processing and reasoning tasks. The
to be installed by an authorized Honeywell dealer. It does centralized gateway usually does not have any interface. They
not work with Wi-Fi enabled smart thermostats. In addition, can be controlled and managed utilizing smartphones, tablets,
the Honeywell provides security cameras and sensors, it also or computers [54], [55].
supports other smart devices from third parties, such as Yale In general, depending on the communication media used,
locks and Lutron lighting. Table IV shows a comparison SB network technology can be classified by interconnection
among various smart control devices in the SBs. method into three main types: Powerline, Busline, and Wire-
less [15] , [56], which we describe next.
C. Networking and Home Gateway 1) Powerline communication (PLC): PLC method reuses
An SB combines a communication network in order to the building electrical network; such that devices, appliances,
control smart devices and services within the building. The and services are directly connected to the main power supply
communication network of a smart building can be based on utilizing the already available electrical outlets in a building.
diverse communication media such as twisted pair cable, as The data is sent through the normal cable system to activate
the traditional computer networks. The networking in building or deactivate the devices in the building. PLC system is
automation system has a tendency to utilize a heterogeneous historically the oldest technology in SB and is generally cheap
network that is made up of diverse communication media and but less reliable and scalable [3]. Originally, the application
network standards. The building automation network is iden- of PLC was mainly to secure the typical operation of the
tified by physical technology and communication protocols. electric power supply system in case of failures or breakdowns
There is an internal network that connects devices inside the through the direct exchange of information between the distri-
building, as well as the external networks, can be integrated bution center, and power plant. Therefore, this approach has
separately. Public Internet, ISDN, and mobile phone networks become a competitive choice for SB networking, benefiting
are some examples of external networks. [53], [28]. from availability, robustness, and ready connectivity of this
A typical SB may comprise a number of different com- method. Some of the protocols of this method offer a single-
ponents, such as sensors, actuators, communication and pro- way communication, which enables the device to only receive
cessing devices. Because of their nature, these components information but not to communicate. There are different main-
have limited capabilities and computational capacity in term stream protocols of PLC method such as X-10, INSTEON,
of battery capacity and capability of data processing. To deal HomePlug, BACnet, and Lonworks.
with this issue, most of the SB systems have been utilized 2) Busline: Busline systems in SBs networks use a separate
as a central gateway to collect, process, and analyze context physical media, usually twisted-pair cabling similar to the
data from different sensors and actuators in the building. physical cables utilized for network services for transporting
Several protocols such as Bluetooth, ZigBee, Wi-Fi, and Z- electrical signals. This type of systems is pleasant the build-
wave can be utilized for communicating the gateway. The ing’s occupants, albeit the configuration process is complex,
and it requires some knowledge of networking. Although the interoperability across for all devices and services in the
configuring complexity and installation cost of this system, building, to simplify the experience for residents. It will enable
the use of a separate cable could present a positive note users to seamlessly and intuitively customize their appliances
about this approach, as it allows this method of networking and devices, regardless of manufacturers and brands of these
to provide higher bandwidth, and to make it the most reliable devices, in order to improve energy efficiency and achieve a
of the three approaches. In addition, this technology usually unique level of control and comfort. For instance, the user
supports a completed two-way communication protocol that can close the blinds in the home either by a click from a
enables the appliances to easily communicate with each other. smartphone or through a pre-set instruction; and switch off
[57]. Some of the protocols in Busline technology are Konnex automatically all screen devices for the children to go outside
(KNX), CAN (Controller Area Network), Modbus, Meter-Bus to play. In a smart building, many devices and appliances can
(M-Bus). simply and securely share information with one another and
3) Wireless interconnection: Many of the new SB appli- with smartphones and other smart devices; and the Internet in
cations use wireless technologies such as infrared and radio general.
frequency, which are more convenient for users due to their Indigo Domotics [60] is to implement the do-it-yourself
untethered nature and the elimination of cables. The devices smart building platform. Indigo home automation software
within the smart building can communicate wirelessly as radio controllers available for the Mac OS enables residents to
wave can penetrate through floors, cabinets, and walls [56]. combine an array of common INSTEON, Z-Wave and X10
Because of the complexity and cost of potential modifications devices for unparalleled control of your building lighting,
and of the re-wiring process in a smart building, several sensors, thermostats, and appliances. With Indigo Touch (sold
different wireless technologies are rising to produce flexible through iTunes app store, iOS only), users will easily achieve
networking patterns convenient to occupants without taking remote control of their appliances utilizing an iPhone, iPad, or
to consider the physical wiring and deployment of such wire iPod Touch. They also can use a web browser on any device to
in the building. Typically, there are various protocols for the control their appliances virtually anywhere in the world. The
wireless system such as Bluetooth, ZigBee, WLAN, Z-wave, users can receive texts or emails about specific events has been
RFID etc., which essentially work in the industrial scientific detected for doors opening/closing, power failures etc. Indigo,
medical bands, particularly in the 2.4GHz frequency range. from Perceptive Automation, is the newest home automation
These wireless technologies are usually related to some control software for the Mac.
network concept in an SB such as low power consumption, OpenHAB [61] is an open-source software platform that
high cost-effectiveness, low speed, flexibility in networking, follows a middleware approach for integrating different tech-
deployment as well as building coverage [3]. The gateway nologies in smart building systems into a single solution.
is the central server of an SB that is commonly used in OpenHAB platform address a variety of network technologies
IoT solutions. The services provided by the gateway essen- and appliances in the area of a smart building. Currently, the
tially concern to system management functionalities such as dependency on a particular vendor becomes a problem due
monitoring, controlling, and configuring the systems and their to the lack of a common language that bridges the different
devices. It also supports some processing and data storage devices with building automation system. The main goal of
capabilities required for complex applications. the OpenHAB platform is to integrate the new devices and
technologies in a smart building system through a community-
D. Software platform based approach. OpenHAB utilizes an OSGi based modular
For a building to be “smart,” it is important that all the appli- system for communicating between different technologies and
ances and systems in the building communicate and exchange devices. Bindings can be developed and deployed as an OSGi
data securely with each other as well as with smartphones, bundle to bridge a particular technology and device. There are
tablets, and servers in the cloud. Software platforms play a different supported technologies exist such as EnOcean, KNX,
critical role in exchanging, archiving and disseminating infor- Z-Wave, and others are supported through special bindings
mation through different protocols. These platforms use push, [62].
pull, publish/subscribe, etc. The goal of the joint commercial SmartThings [47] this platform composed of hardware de-
enterprises is to develop an open source software platform in vices, sensors, and software applications. Context information
order to make the process of data exchanging easier between is collected from the sensors, this context is contributed to the
the devices of different manufacturers. Therefore, the users reasoning and action that are performed by the application. For
will not have to worry in the future about the compatibility instance, the sprinkler in the garden can detect the rain, and
issues when utilizing electric and electronic devices of dif- switch itself off accordingly to save water. SmartThings kit
ferent manufacturers at home. In addition, the new platform comprises sensors, smart devices, and hub. While the Smart-
can also offer a variety of different home services such Things application is configured to enable users to control
as entertainment, energy efficiency, and security technology. and monitor their building environment through a smartphone
Hence, this will enable creating different apps for these areas device. The SmartThings Hub works to connect the sensors,
of use [58]. devices and building’s internet router to one another and
ABB, Robert Bosch GmbH, and Cisco Systems Inc. es- to the cloud. It is compatible with different communication
tablished an open-software platform called Mozaiq Opera- protocols such as Zigbee, Z-Wave, and IP-accessible devices.
tions GmbH [59] to unify smart home technology and offer In addition, the SmartThings is compatible with other sensors
and devices such as thermostats, moisture sensors, motion network simulator. It is distributed with Contiki OS project.
sensors, presence sensors, locks and garage door openers [63]. Cooja enables the emulation of the set of sensor nodes, in
HomeOS [64] is Microsoft’s home operating system plat- addition, it can simulate physical and application layers of the
form, that can be installed on a personal computer. It is an system [71]. There are three basic properties for the simulated
open platform that is not limited to Windows-based devices node in Cooja: Its hardware peripherals, node type, and data
[65] [66]. with HomeOS platform, applications can be installed memory. The node type can be shared among multiple nodes
to maintain various context-aware functionalities, for example, and defines properties that are common to all these nodes [72].
taking an image by a door camera and sending it to the Summary: The field of SBs contains a variety of technolo-
occupant when someone rings the doorbell. HomeOS provides gies, across commercial, industrial, institutional and domestic
a PC-like abstraction that manages and extends the technology buildings, including building controls and energy management
of network devices to the users and developers in the smart systems. Several organizations and institutions are working to
building environment. Its design enables the users to map supply buildings with technology that enables the residents to
their protocol-independent services to support the applications adopt a single device to control all electronic devices and ap-
with simple APIs, a kernel, and protocols of specific devices. pliances. In this section, we discussed the various components
HomeOS usually runs on an allocated computer such as a for SBs including sensors and actuators, smart control devices,
home gateway, it does not need any adjustments to commodity smart gateway, networking and software platforms.
devices. HomeOS usually utilizes (i) Datalog-based access
control to facilitate the process of managing technology in IV. ML B ACKGROUND FOR SB S : M ODELS , T ASKS, AND
the smart home (ii) a kernel to incorporate the devices and TOOLS
applications and (iii) protocol-independent services to allow Massive data generated from sensors, wearable devices, and
the developers manageable access to the devices. other IoT technologies provide rich information about the
Lab of Things (LoT) [67][68] is an experimental research context of users and building status and can be used to design
platform that utilizes connected devices in the buildings. LoT SB management. This context information is needed to extract
offers a framework that provides deployment capabilities such useful and interesting insights for various stakeholders. When
as remote monitoring and updating of system health, and log- the data volume is very high, developing predictive models
ging data collected from different appliances in cloud storage. using traditional approaches does not provide accurate insight
It enables data sharing and collecting, sharing codes, connect and we require newly developed tools from big data. Big data
hardware sensors to the software platform, and participants is primed to make a big impact in SBs and is already playing
using HomeOS. The platform is designed to make it simple a big role in the architecture, engineering, and construction
to design solutions that can be deployed in IoT based smart (AEC) industries [73], notably for waste analytics [74] and
services such as healthcare, energy management services as it waste minimization [75].
works in combination with HomeOS. ML is a powerful tool that facilitates the process of mining a
Eclipse Smarthome [69] is a framework that has a focus on massive amount of data that have been collected from different
heterogeneous environments such as smart building and am- sources around us and make sense of a complicated world.
bient assisted living. This platform takes to consider a variety ML algorithms apply a model on new data by learning the
of existing communication mechanisms. Eclipse SmartHome model from a set of observed data examples called a training
works as an abstraction and translation framework that enables set. For example, after being trained on a set of sample
communications across system and protocol boundaries. It accelerometer data marked as walking or jogging, an ML
provides many relevant implemented extensions, protocols, algorithm can classify the future data points into walking
and standards that are significant for smart building services. and jogging classes. ML makes it relatively easy to develop
Those implementations can be of Java library or an OSGi advanced software systems without much involvement from
bundle shapes so that they can be utilized independently from the human side. They are applicable to many real-life problems
the rest of the project. The framework can work on different in SB environments. One can also design and develop self-
embedded devices such as a BeagleBone Black, an Intel Edi- learning and collaborative systems.
son, or a Raspberry Pi. Extensions of Eclipse SmartHome are ML does not remove the human element from data
compatible with the solutions provided by different vendors. science—it draws on computers’ strengths in handling big data
This means your code that is written for a specific purpose to complement our understanding of semantics and context. It
can be extended easily on commercial platforms. Eclipse only needs training data to extract better features or parameters
SmartHome offers a variety of characteristics to allow you to required to improve a given system. ML algorithms can be
design a special Smart Home solution for your expectations used to make predictions based on data patterns. It enables
[70]. the computer to learn from the fed input data without being
Apart from discussing various SB solutions, we will also explicitly programmed so that ML algorithms can learn from
highlight the popular simulator called Cooja is used widely and make predictions on input data [76][77]. Nest thermostat
by the research community to produce small simulations for is an example of a device that applies a specific temperature in
relatively large wireless networks of embedding sensors and a specific room and at a certain time of day according to the
actuators; and connected devices, in order to develop, debug occupants’ preference. There are devices such as Amazon’s
and evaluate systems based on the wireless sensor network Echo that can learn from voice patterns, and the others those
technology. Cooja simulator is a Java-based wireless sensor learn from much more complex behavior and activity patterns.
A. ML Models a) Classification: The task of classification algorithms is
to classify an instance into a specific discrete set of possible
ML techniques have been widely used to develop smart categories. Given two sets of data (labeled and unlabeled
systems which can sense and react according to context modi- datasets), the labeled dataset is used for the training process,
fications in SBs [78]. There are many different ML algorithms, while the unlabeled dataset will be used to evaluate of the
according to the two well-known theorems No Free Lunch classification results. The normal process is to count the
theorem and Ugly Duckling theorem. No Free Lunch theorem number of instances that are assigned to the right category,
states “there are no algorithms that can be said to be better which is also known as the accuracy rate (AR) defined by
than any other”, without prior information about the problem, [21].
any two algorithms may perform equally well in solving a The classification algorithm can mathematically be de-
problem. While Ugly Duckling theorem states “we cannot say scribed as follows:
that any two different patterns would be more similar to each
other than any other pairs.” [79]. (1)
Mainly, ML is categorized into four categories handling
different types of learning tasks as follows: Supervised learn- where Nc denotes the number of test instances that are
ing, unsupervised learning, semi-supervised learning and re- correctly assigned to their categories to which they belong; Nt
inforcement learning (RL) algorithms Figure 7 shows ML the number of test instances. The precision (P) and recall (R)
styles. These categories are described next and a summarized are used to measure the details of the classification results.
comparison between these ML techniques is presented in Table The four possible outcomes are true positive (TP), false
V. negative (FN), false positive (FP), and true negative (TN), the
precision (P) and recall (R) are generally defined as:
1) Supervised Learning: refers to developing algorithms
based on a labeled training dataset, from which the learner (2)
should generalize a representation by building the system
model that represents the relations between the input, output Given P and R, a simple method to describe the precision
and system parameters. ML model is developed through a and recall of the overall classification results, called F-score
training process that continues on the input training data or F-measure, is defined as:
until the model reaches the desired level of accuracy [80],
[81]. Some examples of common supervised ML algorithms
are: naive Bayes model, decision tree, linear discriminant (3)
functions such as support vector machines (SVMs), artificial
neural networks (ANNs), hidden Markov models (HMMs), Commonly used classification techniques include decision
instance-based learning (such as k-nearest-neighbor learning), trees, SVM, rule-based induction, neural networks, deep learn-
ensembles (bagging, boosting, random forest), logistic regres- ing, memory-based reasoning, and Bayesian networks [89].
sion, genetic algorithms, and logistic regression [82] [83].
b) Decision Tree Algorithms: The decision tree method
Supervised learning approaches are extensively used to solve
is an important predictive ML modeling approach, which
different problems in smart buildings.
constructs a model of decisions presented based on the actual
Application in SBs: Boger et al. [76] proposed a supervised values of features in the data. Decision trees can be utilized for
learning system using Markov decision processes to help both classification and regression problems. In tree structures,
people with dementia the process of hand washing. Altun et leaves represent class labels and branches represent conjunc-
al. [84] make a comparative study on the supervised human tions of attributes that drive to those labels. [90].
activity classification approaches using body-worn miniature The decision trees that the target variable takes continuous
inertial and magnetic sensors. Mozer [85] developed the
values called regression trees. Decision trees are often one
occupant comfort control of the home environment system of the favorites of ML algorithms because of its speed and
using neural networks and reinforcement learning to control air accuracy. The most common algorithms for decision tree are
heating, lighting, ventilation, and water heating in the smart
[91]: classification and regression tree, ID3, C4.5 and C5.0,
home environment. Bourobou et al. [86] presented a hybrid Chi-squared, M5, and conditional decision trees.
approach using ANN and K-pattern clustering to identify and
Application in SBs: Delgado et al. [92] propose an ML
predict user activities in the smart environments. Hsu et al. [87]
technique based on decision trees to extract the most frequent
proposed a TV recommendation system using a neural network
activities of human behavior and the temporal relationship of
model based on user personalized properties such as activities,
those activities in order to produce the human behavior quickly
interests, moods, experiences, and demographic information
in a smart environment. Viswanathan et al. [93] introduce a
data. Fleury et al. [88] proposed a healthcare-focused smart
prototype distributed data mining system for healthcare envi-
home system using the SVM algorithm to classify daily living
ronment using C4.5 classification algorithm that can provide
activities based on the data from the different sensors.
the patient monitoring and health services. Decision trees
Supervised learning problems can be further grouped into algorithm is a non-parametric algorithm that is easy to interpret
classification, regression, time series, and ensemble method and explain. The main disadvantage of this algorithm is that
problems. it can easily overfit.
(a) Supervised Learning (b) Unsupervised Learning
c) Bayesian Algorithms: Bayesian methods utilize SVM model to predict the times where visitors are present
Bayes’ theorem for classification and regression problems. The in the home using only the data provided by wireless motion
most common Bayesian algorithms are [94]: Naive Bayes, sensors in each room. Fleury et al. [88] presented a study
Gaussian naive Bayes, Bayesian belief network, Bayesian for automatic recognition of daily living activities in a smart
network. home based on SVM. They collected the data from various
Application in SBs: Parnandi et al. [95] propose an indoor sensors such as Infra-Red Presence Sensors, door contacts,
localization approach based on Naive Bayes classification and temperature and hygrometry sensor, and microphones. Das et
dynamic time warping, they exploit the embedded sensors of al. [108] proposed a one-class classification approach for a
smartphones to determine the building that the user entered real-time activity error detection in smart homes using one-
and the activities that the user is performing inside the build- class SVM. Zhao et al. [143] proposed an ML approach
ing. Verbert et al. [96] proposed an ML approach based on based on SVM and RNN to detect the occupancy behavior
Bayesian network to diagnosis the fault in HVAC systems. of a building through the temperature and heating source
The model has been constructed based expert knowledge information for the energy efficiency consumption purposes.
concerning conservation laws, component interdependencies, e) Artificial Neural Network Algorithms (ANNs): ANN
and historical data using virtual sensors. Naive Bayes classifier models are inspired by the process of biological neural net-
approaches have been applied with potential results for human works. ANN models are commonly utilized for regression
activity recognition in [97], [98]. Naive Bayes approach recog- and classification problems. The common ANN algorithms are
nizes human activities that identify with the highest probability [94]: Perceptron, Back-Propagation, Hopfield Network, and
to the set of sensor readings that were observed. Radial Basis Function Network (RBFN).
d) Support Vector Machine (SVM): is a supervised ML ANNs provide a number of advantages including it requires
algorithm which can be applied for both classification and less statistical training, it has the capacity to detect complex
regression problems though mostly used in classification chal- nonlinear relationships between the predictor and response
lenges [140]. SVM is one of the most popularly utilized for variables, as well as the ability to detect all possible re-
many statistical learning problems, such as face and object lationships between predictor variables [104]. On the other
recognition, text classification, spam detections, handwriting hand, disadvantages include its “black box” nature, heavy
analysis etc. [141]. is maximizing the margin that separating computational burden and proneness to overfitting. However,
between the hyperplane of two classes’ closest points. Support due to the inherent features of neural networks, it has the
vectors are the points lying on the boundaries, and the optimal following main limitations: challenge in training with no local
separating hyperplane is the middle of the margin [142]. optima, its accommodation to modifications in the behavior,
Application in SBs: Fu et al. [105] proposed an SVM the validation process of the results, and the complexity of
method to predict the system level electricity loads of public network performance interpretation.
buildings that have electricity sub-metering systems. A real- Application in SBs: Badlani and Bhanot [99] developed
time human tracker system proposed Nguyen et al. [106] using a smart home system for energy efficiency applying pattern
SVM for predicting and recognizing human motion based recognition based on ANNs, the system incorporates an RNN
on the input images from a network of four cameras in the to capture human behavior patterns and an ANN for security
ubiquitous smart homes. Petersen et al. [107] developed an applications in smart homes. Other researchers have applied
TABLE V: COMPARISON OF ML TECHNIQUES
Category Type Algorithms Pros Cons Applicability in SBs Cited
computational burden; Prone to
Requires little statistical train- Used for classification, control and [99][100]
Neural Overfitting; Picking the correct
ing; Can detect complex non- automated home appliances, next [101][102]
Classifi- networks topology is difficult; Training can
linear relationships step/action prediction. [103][104]
cation take a long time and a lot of data;
Can avoid overfitting using
Computationally expensive; Classification and regression prob-
Supervised the regularization; expert [105][106]
Slow; Choice of kernel models lems in SBs such as activity recog-
Learning SVM knowledge using appropriate [107][88]
and parameters sensitive to nitions, human tracking, energy ef-
kernels [108]
overfitting ficiency services
Very simple representation
Bayesian net- does not allow for rich hy- You should train a large training Energy management system and hu- [109][96]
works potheses set to use it well. man activity recognition. [97][98]
ANNs to present context-aware services. Campo et al. [100] ture extraction algorithms. The most common DL algorithms
developed a system that calculates the probability of occu- are [145]: Convolutional Neural Network (CNN), Recurrent
pation for each section of the building and compares the Neural Network (RNN), Deep Boltzmann Machine (DBM),
probability with the current situation systematically. See [101] Deep Belief Network (DBN), and Stacked Auto-Encoders.
for a survey paper focusing on the role of ANNs for smart Deep learning has been used successfully in varieties of big
home services. Ermes et al. [102] proposed a hybrid classifier data analytics applications, particularly natural language pro-
approach using a tree structure comprising a priori knowl- cessing (NLP) applications, medical diagnosis, stock market
edge and ANN to recognize the activities such as rowing, trading, network security, and image identification.
biking, playing football, walking, running, sitting, or hiking.
Ciabattoni et al. [103] proposed a home energy management Deep learning is now ubiquitously used in major businesses
system design using the neural network algorithm to predict and companies. Microsoft research on a deep learning system
the power production of the photovoltaic plant and the home presented real-time speech translation system between Man-
consumptions during the given time. darin Chinese and English languages [146]. Apple’s Siri uses
a deep learning trained model, and the voice recognition in
f) Deep Learning Algorithms: Deep learning methods the Google Android phone also uses a deep learning trained
represent an evolved form of ANNs in which a deep architec- model [147]. DL utilizes a number of techniques such as drop-
ture (many layers comprising multiple linear and non-linear out and convolutions that enables the models to learn
transformations [144]) is used. One of the promises of DL is efficiently from high-dimensional data. However, DL requires
replacing the manually selected features with efficient unsuper- much more data to train compared to other algorithms because
vised or semi-supervised feature learning and hierarchical fea- of the magnitudes of parameters for estimation required by the
models. Application in SBs: Survadevara et al. [125] proposed
Application in SBs: Choi et al. [114] propose two predic- a wellness model using seasonal autoregression integration
tion algorithms deep belief network and restricted Boltzmann moving average time series with sleeping activity scenario
machines based on the DL framework for predicting different in a smart home environment to forecast the elderly sleeping
human activities in a building. They also presented a hybrid tendency. Zhou et al. [126] proposed a time series analysis
model which combines for predicting human behavior. The framework to explore relationships among non-stationary time
paper [115] proposes a generic deep learning framework based series in the case of data sensors in SBs. Jakkula and Cook
on convolutional and RNNs for human activity recognition [127] propose a time series based framework to determine tem-
that is suitable for multimodal wearable sensors, such as ac- poral rules from observed physical and instrumental activities
celerometers, gyroscopes or magnetic field sensors. Alsheikh of occupants in a smart home.
et al. [116] proposed a hybrid approach of DL and hidden i) Regression: The aim in regression problems is to
Markov model for human activity recognition using triaxial estimate a real-valued target function. It is related to repre-
accelerometers. Baccouche et al. [117] propose a two-steps senting the relationship between variables that are repeatedly
neural-based deep model to classify human activities, the first processed utilizing a measure of error in the predictions made
step of the model is automatically learned spatiotemporal by the model [152]. The most common regression algorithms
features based on Convolutional Neural Networks. Then the are [153]: linear regression, logistic regression, stepwise re-
second step of the model uses an RNN to classify the entire gression, and ordinary least squares regression.
sequence of the learned features for each time-step. In [118], Application in SBs: Chen et al. [120] used the regression
they propose an acceleration-based human activity recognition technique of orthogonal matching pursuit algorithm to identify
method using Convolution Neural Network. In [119] a deep the physical and environmental parameters that providing the
convolutional neural network as the automatic feature extractor energy efficiency in an SB. Bouchard et al. [121] presented a
and classifier for recognizing human activities is proposed gesture recognition system using linear regression combined
using the accelerometer and gyroscope on a smartphone. with the correlation coefficient to recognize the gesture direc-
Hammerla et al. [148] explore the performance of deep, tion and estimate the segmentation of continuing gestures of
convolutional, and recurrent approaches of deep learning for daily usage activities in a smart environment.
human activity recognition using wearable sensors. For the
j) Ensemble methods: A combination of multiple classi-
sake of measuring the performance, the authors used three
fiers often referred to as a classifier ensemble, group of classi-
representative datasets that comprise motion data collected
fication models that are trained separately and the predictions
from wearable sensors.
of those models are then combined in a way to produce the
g) Hidden Markov models (HMM): An HMM is a dou- overall prediction [154]. The most popular ensemble learn-
bly stochastic process with a hidden underlying stochastic ing based classification techniques are [155]: random forest,
process that can be observed through the sequence of observed boosting, gradient boosting machines, AdaBoost, bagging, and
symbols emitted by another stochastic process. blending.
Application in SBs: Wu et al. [113] proposed an improved Application in SBs: Jurek et al. [122] proposed a cluster-
HMM to predict user behaviors in order to provide services based ensemble approach solution for activity recognition
for people with disabilities. They developed a temporal state within the application domain of smart homes. With this
transition matrix to be utilized instead of the fixed state approach, activities are modeled as cluster collections built on
transition matrix. Lv and Nevatia [112] used hidden Markov different subsets of features. Fatima et al. [123] proposed an
models for both automatic recognition and segmentation of 3- ensemble classifier method for activity recognition in smart
D human activities to allow real-time evaluation and feedback homes using genetic algorithm optimization to merge the
for physical rehabilitation. Cheng et al. [110] proposed an prediction output of multiple classifiers that make up the en-
inference engine based on the HMM that provides a com- semble. They used the ANN, HMM, conditional random field,
prehensive activity of daily living recognition capability. They and SVM [13] as base classifiers for activity recognition. Guan
integrated both Viterbi and BaumWelch algorithms to enhance and Ploetz [124] proposed a deep LSTM ensemble method
the accuracy and learning capability. Chahuara et al. [111] for activity recognition using wearables: more specifically, the
proposed sequence-based models for online recognition of authors developed modified training procedures for LSTM
daily living activities in an SB environment. They presented networks and proposed the combination of sets of diverse
three of sequence-based models: HMM, conditional random LSTM learners into classifier collectives.
fields, and a sequential Markov logic network. 2) Unsupervised Learning: Unsupervised Learning refers
h) Time Series Analysis: A time series is a collection to developing algorithms that use data with no labels to
of temporal instances; time series data set usually have the analyze the behavior or the system being investigated [156].
following characteristics include the high dimensionality, large Thus, the algorithm does not know about the truth of the
number of instances, and updating continuously [149]. One outcome. In other words, the unsupervised learning algorithm
of the important purposes for time series representation is classifies the sample sets to different clusters by investigating
to reduce the dimension, and it divides into three categories: the similarity between the input samples. Clustering is done
model-based representation, non-data-adaptive representation, using different parameters taken from the data which enable us
and data-adaptive representation [150] [151]. to identify correlations which are not so obvious. The inferring
structures existing within the input data is used to prepare the home environment based on a Markov modulated Poisson
model to prepare and extract general rules of the model. A process model. The model combines multiple data streams,
mathematical process might be used to systematically reduce such as in the front-door sensor transitions and the general
redundancy, or organize data by similarity [129]. sensor transitions. The other cases of social communication
The unsupervised approach has been applied to recognize services, Cook et al. [129] applied an unsupervised learning
various activities in smart buildings when it is challenging approach to detect social interaction and monitor activity daily
to have labels for input data [130]. Common unsupervised living in a smart space, their approach can adapt and update
learning problems are clustering, dimensionality reduction, automatically to reflect the changes in discovered patterns
and association rule learning. There are a variety of com- from implicit and explicit identified feedbacks of the occupant.
monly used unsupervised learning algorithms, some of those Rashidi et al. [130] introduce an unsupervised method that
algorithms are based on supervised-learning algorithms: the identifies and tracks the normal activities that commonly occur
Apriori algorithm and k-Means. In unsupervised learning, in an individual’s routine in a smart environment. The activity
usually there is no a measure for the output; we recognize discovery method of the system is produced to cluster the
only the features and the target is to define the patterns and sequences based on the simple k-means algorithm. Fiorini et
relationships among a set of input measures [80]. al. [159] proposed an unsupervised ML approach to identify
The major disadvantage of unsupervised learning is the the behavioral patterns of the occupants using unannotated
absence of direction for the learning algorithm, hence, there data collected from low-level sensors in an SB. Their approach
might not be any useful detected knowledge in the selected involves processing and analyzing collected data related to the
set of attributes for the training. Clustering is a method daily living activities of 17 older adults living in a community-
of unsupervised learning that involves detecting patterns in based home supplied with a variety of sensors. They extract
the data by placing each data element into a group of K- activity information from collected data at different times of
clusters, where each group holds data elements most similar the day.
to each other [157]. Unsupervised learning problems can be b) Association: The association rule learning problem is
categorized into clustering and association problems, which utilized to identify the rules that define large portions of input
are described next. data, such as people that buy X item also tend to buy Y
a) Clustering: A clustering problem explores the internal item. Association analysis is performed on rules discovered by
groupings in the input data, such as grouping customers analyzing input data for frequent if/then statement and using
by their purchasing habits. Clustering techniques are usually the criteria of support and confidence to discover relationships
organized by modeling strategies such as centroid-based and between unrelated data in a relational database or another in-
hierarchical. All methods are concerned with handling the formation repository. Here “support” indicates how frequently
internal structures in the input data to properly organize the items appear in the database while “confidence” indicates
the data into groups of maximum commonality [158]. The the number of times the if/then statements have been found
quality of the clustering result is evaluated depends on the to be true. Many algorithms for generating association rules
type of application that utilizes a clustering algorithm. For have been proposed. Apriori algorithm is the most well-known
example, the sum of squared errors is generally utilized for association algorithm [160].
data clustering while the peak-signal-to-noise ratio is used for Application in SBs: Aztiria et al. [161] proposed system
image clustering [21]. The most common clustering algorithms that learns the frequent patterns of human behavior using
are [153]: k-Means, k-Medians, expectation maximization, and association, workflow mining, clustering, and classification
hierarchical clustering. techniques. The core part of the system is the learning layer
which is made up of two modules: the language module,
Application in SBs: Fahad et al. [128] propose an activity
which provides a standard conceptualization of the patterns;
recognition approach that combines the classification with
and the algorithm module, which discovers the patterns. Kang
the clustering, in their approach the activity instances are
et al. [162] proposed a service scenario generation scheme
clustered using Lloyd’s clustering algorithm. Then, they apply
for interpreting association rules extracted from the states
evidence theoretic K-Nearest neighbors learning method that
of all devices in SB environments. Typically, These states
combines KNN with the Dampster Shafer theory of evidence.
are collected periodically at a specific time interval from
The paper [86] proposes a hybrid approach to recognize
the devices. Nazerfard et al. [163] propose a framework to
and predict user activities in a smart environment. They use
discover the temporal features of the activities, including the
the K-pattern clustering algorithm to classify so varied and
temporal sequencing of activities and their start time and
complex user activities, and ANN to recognize and predict
duration using the temporal association rule techniques in a
users’ activities inside their personal rooms. Lapalu et al.
smart home.
[82] used an unsupervised learning approach to address the
issues of daily living activities’ learning in smart home. They 3) Semi-Supervised Learning: Semi-Supervised learning
utilize the Flocking algorithm for clustering analysis of a use lies between supervised and unsupervised methods. Input
case in cognitive assistance service that assists the people data is a composite of labeled and unlabeled samples. These
suffering from some type of dementia such as Alzheimer’s hybrid algorithms aim to inherit the strengths of the main
disease. Aicha et al. [83] present an unsupervised learning categories while mitigating their weaknesses. The model learns
approach for detecting abnormal visits of an elderly in a smart the patterns present in the data and also make predictions.
Example problems are classification and regression [164]. method for autonomous learning of a user’s preference of
There are some common semi-supervised learning mod- music and lighting service settings in presence of different
els, including generative models, heuristic approaches, semi- states of the user in SB environment. The preferences are
supervised SVM, graph-based methods, self-training, help- learned by the model by using the explicit or implicit feedback
training, mixture models, co-training and multi-view learning from users when they react to the provided service. Xu et
[94]. al. [139] give a survey of developments in RL algorithms
Application in SBs: Cook [131] combined fully-supervised with function approximation. They evaluated and compared
and semi-supervised learning to recognize and follow activities different RL algorithms using several benchmark learning,
that support health monitoring and assistance context-aware prediction, and learning control tasks.
services for people experiencing difficulties living individually
at smart homes. Liu et al. [132] proposed a vision based semi- B. ML Tasks for SBs
supervised learning approach for fall detection and recognizing In this section, we will describe the major ML tasks that
other activity daily living in smart environments to overcome are relevant for SB. The reader is referred to Figure 8 for a
the labeling challenges of human activities by systematic general depiction of ML tasks in SBs and the steps taken to
interpreting the activities with the highest confidence. Fahmi implement ML in an SB environment.
et al. [133] proposed a semi-supervised fall detection approach
in which a supervised algorithm utilizing decision trees in the 1) Data Collection and Acquisition: A variety of data
training process and then profiles are used to implement a collection approaches are used, each of which has different
semi-supervised algorithm based on multiple thresholds. Radu deals in terms of capabilities, energy efficiency, and con-
et al. [134] present semi-supervised ML method using only the nectivity. Sensors and similar objects in SBs produce raw
low power sensors on a smartphone to consider the problem data simultaneously in an automated way and such devices
of determining whether a user is indoors or outdoors. Guan may store the data for a specific period of time or report
et al. [135] propose a semi-supervised learning algorithm for it to controlled components [168]. Data can be collected at
activity recognition named En-Co-training to make use of the gateways; the collected data is then filtered and processed,
available unlabeled samples to enhance the performance of fused into compact forms for efficient transmission. A variety
activity learning with a limited number of labeled samples. of communication technologies such as Zigbee, Wi-Fi, and
The proposed algorithm extends the co-training paradigm by cellular are utilized to transfer data to collection points.
using an ensemble method. Data collected from a global-scale deployment of smart
things defines the basis for decision making and providing
4) Reinforcement Learning: Reinforcement learning is a services. It is possible that the decisions are unreliable when
learning approach to control a system in order to maximize the quality of utilized data is poor [169].
performance measure that represents a long-term objective Zhao et al. [170] propose a data acquisition and transmission
[165]. Reinforcement learning, an area of ML inspired by system which could be used for monitoring systems to collect
behaviorist psychology, is concerned with the way that soft- energy consumption data (e.g., electricity, water, gas, heating,
ware agents have to take actions in an environment in order etc.) from terminal meters which are installed in buildings.
to maximize the concept of cumulative reward. RL algorithms The system stores the data periodically after analyzing and
learn control policies, particularly when there is no a priori processing it and finally transmits the data to servers through
knowledge and there is a massive amount of training data. the Ethernet. Rowley et al. [171] propose the data acquisi-
However, RL algorithms suffer from some drawback such tion and modeling approaches that can support the delivery
as the high computational cost required to find the optimal of building energy infrastructure in both new building and
solution, such that all states need to be visited to choose the renovated real-world contexts. Such methods provide a means
optimal one. The well-known approaches of RL are Brute to achieve short, medium and long-term forecasting of possible
force, Monte Carlo methods, Temporal difference methods, scenario pathways to multi-objective sustainable outcomes.
and Value function [166]. Q-learning [167] is a model-free re- CLEEN MMEA [172] platform that collects, processes, and
inforcement learning approach based on learning the required manages the data and initiates contextual knowledge extrac-
utility given a state decision. tion. The purpose is to establish an online marketplace to
Application in SBs: Mozer [136] applied Q-learning for collect data and provide services for different companies. The
lighting regulation to predict the time of turning the lights interfaces are made public so that any company can easily
ON/OFF in a building. This prediction model can be utilized join the network to buy or sell services. The analysis results
to schedule the lights’ activations in a building for efficient can be given to an energy services company in order to allow
energy consumption proposes. Li and Jayaweera [137] pro- offering the service to the owners.
posed a Q-learning based approximate dynamic programming A typical example of open access data collection system
algorithm to provide a more efficient, flexible and adaptive is e3Portal [173] developed by VTT in collaboration with
method. This approach can enable customers to make an Finnish municipalities. e3Portal offers information and tools
optimal on-line decision making in SB environment to max- when planning savings measures and energy retrofitting in
imize the profits based on both local fully observable and municipal buildings. It also involves frequently updated data
the estimated hidden information of the building. Khalili and regarding energy and water consumption in thousands of
Aghajan [138] proposed a temporal differential class of RL public buildings like schools, kindergartens, offices, hospitals,
Fig. 8: ML tasks in SB environment.
other health care facilities, etc. Decision makers, designers, addition, the analog sensors are installed to support readings
operation and maintenance personnel, as well as buildings for hot water, cold water, and stove burner use [179]. The
users, can utilize it. PlaceLab project [180] of MIT provides a dataset collected
from a one-bedroom apartment with more than 900 sensors,
There are projects that provide publicly available SB including those coming from motion, switch and RFID sen-
datasets for researchers to conduct further studies; A list of sors. That is being used to monitor activity in the environment
“Home Datasets” [174] includes the datasets collected by in the context of a smart home [181]. A collection of smart
projects from UC Berkeley, MIT, Washington State University, meters data from five houses in the UK [182] That consists
University of Amsterdam, University of Edinburgh, and the of 400 million instances. The active power is formed by
University of Tokyo. The WARD [175] project supported by different appliances and the whole-house power demand every
NSF TRUST Center at UC Berkeley provides a benchmark 6 seconds.
dataset for human action recognition using a wearable motion
sensor network. The dataset was collected from 13 repetitive The major challenges that arise for data collection are
actions by 13 male and 7 female participants between the scalability, privacy, security, and heterogeneity of resources
ages of 19 and 75. An MIT project [176] collected daily live [183]. Automated sensor data collection process collects a
activities dataset from two single-person apartments within a large amount of data that overwhelms the collection and
period of two weeks. Eighty-four sensors to record opening- analysis centers in comparison to the data collected from other
closing events were attached to different appliances and de- sources such as IoT devices and social media. This leads
vices such as drawers, refrigerators, containers, etc. Baos et to a huge number of small synchronous write operations to
al. [177] introduced an open benchmark dataset collected from the database storage system, consequently, resulting in serious
various inertial sensors attached to different parts of the body. performance bottlenecks to the storage system design [184].
They considered 33 fitness activities, recorded using 9 inertial Because of the extensive use of RFID technology, privacy
sensor units from 17 participants. The CASAS project [178] issues arise in data collection; for example, the RFID tags
at Washington State University provides a publicly-available carried by a person may become a unique identifier for that
dataset for a three-bedroom apartment with one bathroom, a person. Also, other security concerns appear, for example, the
kitchen, and a living room. Different types of motion and radio signals of RFID technology are easily jammed. Hence,
digital sensors are installed to support temperature readings, in that can disrupt the data collection process [185].
The heterogeneity of data that is being collected from dif- tion. Most of these techniques can be utilized in classification
ferent resources is another major challenge, such that the data and regression problems. Examples of some salient algorithms
are usually very noisy, large-scale, and distributed. This makes are [153]: Principal Component Analysis (PCA), Principal
it very difficult to use the collected data effectively without a Component Regression (PCR), and Linear Discriminant Anal-
clear description of existing data processing techniques [184]. ysis (LDA). Chen et al. [191] propose a framework using the
2) Data Preprocessing: A large amount of data are gen- classification information of local geometry of data to reduce
erated by sensors in SBs; this data comes from various the dimensionality of a dataset on human activity recognition
sources with diverse formats and structures. Usually, this from wearable, object, and ambient sensors.
data is not ready for analysis as it might be incomplete or a) Feature Extraction: The main components of the
redundant due to low battery power, poor calibration, exposure original data are the features. After extracting the features from
to various malicious elements and interference. Therefore, the raw dataset, such features contain important information
raw data typically needs to be preprocessed to deal with that is used by the learning algorithms for the activities dis-
missing data, discard noisy and redundant data and integrate crimination. The most common methods of feature extraction
data from various sources into an integrated schema before work in time, frequency, and discrete domains [192]. Among
being committed to storage. This preprocessing is called data time domain method, mean and standard deviation are the key
cleaning. The quality of data can be improved substantially approaches for almost all sensor types. While the frequency
by applying some cleaning techniques to the data before it domain method focuses on the periodic structure of the col-
arrives its end user [168][186]. Data cleaning is one of the lected data. Wavelet Transformation and Fourier Transform are
significant tasks in the data processing phase. Data cleaning the most common approaches. And discrete domain methods
is not a new process particular for the IoT data processing. It such as Euclidean-based distances, dynamic time warping, and
has already been applied as a process for database management Levenshtein edit distance are key approaches implemented in
systems. Presenting a data cleaning method would further aid several applications such to string similarity, classifying human
the applications to focus on their core logic without worrying activities and modeling human behavioral patterns [16], [193].
about data reliability post-processing overheads [184]. b) Feature Selection: The main role of feature selection
There are many different techniques that have been utilized is to discriminate the most related subset of features within a
to deal with the problem of cleaning noisy data streams such high dimensional vector of features, so that reduces the load
as Kalman filters [187], statistical models [186] and outlier of noise and computational expense on the learning models.
detection models [184]. One of the major challenges with data In order to map the high dimensional vector of features
cleaning techniques in the SBs is the heterogeneity of data into a lower dimensional vector, there are several common
collected from different sources particularly WSN- and RFID- algorithms used such as Linear Discriminant Analysis (LDA),
enabled data streams. The utilized data cleaning techniques Principal Component Analysis (PCA), and Independent Com-
should be able to deal with several different variables of ponent Analysis (ICA) [194].
interest to satisfy IoT applications’ requirements, for example, Hausmann and Ziekow [195] proposed an approach for
setting home temperature based on observed outer tempera- automatically adapting the feature selection for SBs applica-
ture, user habits, energy management, etc. [169] Any type of tion ML models from the time-series data based on wrapper
failures such as a failed sensor, network issues, camera failure, methods and genetic optimization. Fahad et al. [196] propose
or database crashes in the process of collecting data would an activity recognition approach for overlapping activities us-
invalidate the data. Consequently, this type of impediment will ing K-Nearest neighbors approach that distinguishes the most
dramatically increase the time required to collect data [179]. important features from the collected information obtained
3) Dimensionality Reduction: There are huge volumes of from deployed sensors in multiple locations and objects. Fang
raw data that are captured from heterogeneous and ubiquitous et al. [197] determine that the different feature sets generate
of sensors used in SBs. Most of the data collected from those different levels of accuracy for recognizing human activities,
sensors are redundant and they need to be brought down to a and selecting inappropriate datasets increases the level of
smaller number of features by applying dimensionality reduc- computational complexity and decreases the level of prediction
tion techniques without losing significant information [188]. accuracy in smart home environments. The wrapper and filter-
The main idea from the dimensionality reduction strategy ing are the two main statistical methods of feature selection
is to find a new coordinate system in which the input data problem. It is argued that although the wrapper approach may
can be represented with much fewer features without losing obtain better performances, filters are less resource intensive
significant information. The dimensionality reduction can be and faster [198]. In [199], different feature selection methods
made in two different ways: by extracting of the features that are utilized for the process of dimensionality reduction of
represent the significant data characteristics (this technique the learning problem to recognize the human activities from
is called feature extraction), or by only selecting the most observed sensors. The authors show that the performance of
relevant features from the original dataset, this method is the learning models to recognize the human activity has a
called feature selection [189] [190]. strong relationship with the utilized features.
Like clustering methods, the dimensionality reduction ap- c) Feature projection: feature projection can be repre-
proach explores and exploits the internal structure of the data, sented as a mapping from the original set of features to
but in this case in an unsupervised manner using less informa- an appropriate set that optimizes the learning criterion, such
that the feature projection approach allows the process of 1) H2O: H20 [203] is an open-source in-memory, dis-
visualizing and mapping the high-dimensional feature vectors tributed, and scalable ML framework for big-data analysis that
to low dimensional one, in addition, it enables analyzing supports ML libraries, along with tools for parallel process-
the distribution of the reduced feature vectors [200]. Conse- ing, analytics, data preprocessing and evaluation tools. It is
quently, the feature projection approach reduces the pattern produced by the H2O.ai, which launched in 2011 in Silicon
recognition’s processing time and enables selecting the best- Valley. The most notable feature of this product is that it
performed classifier for the reduced feature vectors. Hence, provides numerous tools for deep neural networks. The H2O
it makes real-time implementation possible [198]. Chu et al. software APIs can be called from Python, Java, R, and Scala.
[201] proposed a linear supervised feature projection that Users without programming expertise can still utilize this tool
utilizes the LDA algorithm for EMG pattern recognition that via the web-based User Interface. In addition to the processing
attempted to recognize nine kinds of hand motion. engine provided by H2O framework, it also allows the users
to integrate their models with other available frameworks such
C. ML Tools & Platforms for SBs as Spark and Storm. Depending on what is suitable for the
There are a variety of existing ML platforms and tools to algorithm, The H2O’s engine uses multiple execution methods
support the learning process. With the current increasing num- to process data completely in memory. The general technique
ber of those kinds of toolkits, the task of selecting the right tool used is distributed Fork/Join, which is reliable and suitable for
for processing big data streaming from various sources can still massively parallel tasks.
be difficult. Typically, there is no single toolkit that truly fits The H2O software can be run on different operating systems
and provides solutions for all different problems. Many of the such as Microsoft Windows, Mac OS X, and Linux (e.g.
available toolkits might have overlapping uses, and each has Ubuntu 12.04; CentOS), It also runs on Apache Hadoop
advantages and disadvantages. Most of those toolkits might Distributed File System (HDFS) and Spark systems for big-
require experiences in the domains of programming languages data analysis. In addition, it can operate on various cloud com-
and system architecture. In addition, usually many people lack puting environments such as Amazon EC2, Google Compute
a full understanding of the capabilities and how to use those Engine, and Microsoft Azure. As of July 2016, the algorithms
available platforms [202]. supported in H2O cover the tasks classification, clustering,
The important factors that must be considered when se- generalized linear models, statistical analysis, ensembles, op-
lecting a specific ML tool are scalability, speed, coverage, timization tools, data preprocessing options and deep neural
usability, extensibility, and programming languages support. networks.
With respect to the scalability factor, the size and complexity
2) MLlib (Spark): MLlib [204] is Apache Spark’s ML
of the data should be considered to determine if a specific
library. MLlib aims to provide scalable and easy to use ML
toolkit will be fit. The processing platform that the library
methods. It includes common ML algorithms for classification,
is running on and the complexity of the algorithm affect the
regression, clustering, dimensionality reduction, as well as
speed factor. Not all the projects prioritize the speed factor;
lower-level optimization primitives and higher-level pipeline
if the models require frequent updates, the speed may be a
APIs. The classification techniques of SVM, random forest,
crucial concern; but not otherwise. Coverage represents the
logistic regression, Na¨ıve Bayes, decision trees, and gradient-
number of ML algorithms implemented in the tool. With the
boosted trees are supported whereas for clustering, k-means,
massive amount of data capturing from heterogeneous sources,
Gaussian mixture, and power iteration clustering are sup-
ML community faces the challenges of how the ML model can
ported. MLLib supports implementations for linear regression
efficiently process and learn from the big data.
and isotonic regression, and incorporates a collaborative filter-
In general, the available big data tools do not implement all
ing algorithm using alternating least squares. PCA is supported
varieties of different classes of ML algorithms, and typically
for dimensionality reduction. MLlib includes APIs for devel-
their coverage ranges from a few algorithms to around two
opment in Scala, Java, Python, and SparkR. Generally, MLlib
dozen. The usability factor includes elements such as initial
depends on Spark’s iterative batch and streaming approaches,
setup processing; continuous maintenance; the available pro-
as well as its use of in-memory computation.
gramming languages and user interface available; the amount
of documentation, or availability of a knowledgeable user. The 3) TensorFlow: Tensorflow [205] is an open source soft-
extensibility factor means that the implementations introduced ware library for numerical computation and deep ML in a
in the tools can be utilized as building blocks towards new variety of perceptual and language understanding tasks utiliz-
systems. It is necessary to evaluate tools in terms of how well ing data flow graphs. TensorFlow was originally developed by
they are able to meet this factor. There are a variety of ML the Google Brain team and was released in November 2015
libraries that are available in different programming languages. under an Apache 2.0 open source license. TensorFlow has tools
Depending on the task you are trying to accomplish, certain that support deep learning, reinforcement learning, and other
languages, libraries, and tools can be more effective than algorithms. TensorFlow implements data flow graphs, where
others. The following provides a detailed observation of the “tensors” are batches of data that can be processed by a set
strengths and weaknesses of the top used deep learning and of algorithms defined by a graph.
ML tools. The reader is also referred to Table VI for a concise The movements of the data through the system are called
tabulated summary of the described deep learning and ML “flows”—hence, the name. TensorFlow can run on multiple
tools. CPUs and GPUs. It can run on Linux, Mac OS X desktop,
and server systems, and Windows support on roadmap, as well proved by users in other fields such as robotics, neuroscience,
as on Android and Apple’s iOS mobile computing platforms. speech recognition, and astronomy. In addition, it supports
TensorFlow is written with a Python API over a C/C++ engine Python and MATLAB code bindings. Caffe offers image
that makes it run fast. TensorFlow utilizes a symbolic graph classification with state of the art CNN algorithm. Caffe is
of vector operations method, in order to easily define a new mainly utilized as a source of pre-trained models hosted on
network. However, TensorFlow has a weakness that is related its Model Zoo site.
to modeling flexibility. Such that each computational flow Caffe is useful for performing image analysis using CNNs
has to be constructed as a static graph. That makes some and regional analysis within images using RCNNs. The per-
computations such as beam search difficult. formance and processing speed of Cafee make it as one
4) Torch: Torch [206] is an open source ML computing of the most utilized platforms for research experiments and
framework that supports a variety of ML algorithms. Torch industry deployment. It has the capability to process over 60
was originally developed at NYU. It is efficient and easy to million images per day with a single NVIDIA K40 GPU.
use, thanks to a script language based on the Lua programming Caffe has already been applied in many research projects at
language and a C/CUDA implementation, Torch was intended UC Berkeley and other universities, performing very well in
to be portable, fast, extensible, and easy to use in development. many tasks such as object classification, object detection, and
Some version of Torch is employed by large companies such Learning Semantic Features. It provides a complete and well-
as Google DeepMind, the Facebook AI Research Group, IBM, documented toolkit for training, testing, tuning, and deploying
Yandex, and the Idiap Research Institute. In addition, it has models. Caffe utilizes a large repository of pre-trained neural
been extended to run on Android and iOS platforms. A variety network models called the Model Zoo, which is suitable for a
of community-contributed packages for Torch, giving it a variety of common image classification tasks [225].
versatile range of support and functionality. It provides various 8) Azure ML: Microsoft first launched Azure ML [226]
deep learning algorithms that support computer vision; signal, as a preview in June 2014. Azure ML enables users to
image, video, and audio processing; parallel processing and create and train models, then convert those models into APIs
networking [207]. that can be applied to other services. Users can get up to
5) Deeplearning4j: Deeplearning4j [221] is an open source 10GB of storage per account for model data, although they
distributed DL library, primarily developed by Adam Gibson can also connect their own Azure storage to the service for
from an ML group in San Francisco. Deeplearning4j is written larger models. programmers can use either the R or Python
for Java and JVM as well as to support a variety of DL programming language for developing with Azure services.
algorithms such as restricted Boltzmann machine, deep belief Users can purchase ML algorithms from Microsoft Azure
networks, convolutional networks, recurrent neural networks, Marketplace, they can also obtain free algorithms from the
deep autoencoder, stacked denoising autoencoder, and re- community gallery that has been created by Microsoft to
cursive neural tensor network. All these algorithms can be share ML algorithms with each other. They share many of
integrated with Hadoop and Spark for distributed parallel pro- predictive analytics of personal assistant in Windows Phone
cessing. Deeplearning4j relies on Java programming language, called Cortana. Azure ML also utilizes solutions from Xbox
in addition, it is compatible with Clojure and includes a Scala and Bing.
API. Deeplearning4j is designed to be utilized in business Azure currently supports different features and capabilities
environments, rather than as a research tool. It is applied such as run Hadoop over Ubuntu Linux on Azure, it also
in a variety of applications such as fraud detection, anomaly supports hosting Storm for analyzing data streams. In addition,
detection, recommender systems, and image recognition. it allows developers to connect .NET and Java libraries to
Storm. Azure ML studio supports a variety of modules for
6) Massive Online Analysis (MOA): MOA [222] is one of
training, scoring, and validation processes. Azure ML comes
the common open source frameworks for data stream mining
with a large library of algorithms for predictive analytics.
and possessing. MOA is written in Java related to the WEKA
The popular families of algorithms are regression, anomaly
project that developed at the University of Waikato, New
detection, clustering, and classification.
Zealand. It includes a set of learners and stream generators
that can be used from the GUI, the command-line, and the
D. Real-time Big Data Analytics Tools for SBs
Java API. MOA supports a variety of ML algorithms for
classification, regression, clustering, outlier detection, as well Several applications need to have real-time data analysis for
as some tools for evaluation [223]. stream data and waiting for the information to be archived and
then analyzed is not practical for these type of applications.
7) Caffe: Caffe [224] is a DL framework, it is primarily
Generally, Stream processing is intended to analyze a massive
developed with the consideration of expression, speed, and
modularity. It utilizes the machine-vision library for fast amount of data and act on real-time streaming data utiliz-
convolutional networks from Matlab, which has been ported ing continuous queries such as SQL-type queries to handle
to C and C++. It is developed by the Berkeley vision and streaming data in real-time utilizing scalable, available and
learning center and by the community contributors. In Caffe, fault-tolerant architecture. Essential to stream processing is
multimedia scientists and practitioners have an organized and Streaming Analytics. More and more tools offer the possibility
state-of-the-art toolkit for DL algorithms. Caffe was originally of real-time streaming data. The following presents some of
developed for machine-vision, it has been utilized and im- the common and widely used options.
TABLE VI: COMPARISON BETWEEN DEEP LEARNING AND ML TOOLS
CUDA
Open Written Release
Tool Creator OS Interface sup- Algorithms Used in
source? In date
port?
Linux, Mac OS X deep learning algorithm:
Google C++, November
TensorFlow (Windows support on Yes Python, C/C++ Yes RNN, CN, RBM and [208][209]
Brain team Python 2015
roadmap DBN
deep learning algorithm:
Universit de September [210][211]
Theano Cross-platform Yes Python Python Yes RNN, CN, RBM and
Montral 2007 [212]
DBN
Linux, Mac OS, Algorithms for classifica-
Microsoft Windows tion, clustering, general-
And Cross-platform incl. Java, ized linear models, statis-
August
H2O H2O.ai Apache HDFS; Amazon Yes Scala, Python, R No tical analysis, ensembles, [213]
2011
EC2, Google Compute Python, R optimization tools, data
Engine, and Microsoft preprocessing options and
Azure. deep neural networks.
Various,
Linux, OSX, Windows, Java, Deep learning algorithms
Original Java, Scala, Clo- August
Deeplearning4j Android, CyanogenMod Yes Scala, C, Yes including: RBM, DBN, [214]
author Adam jure 2013
(Cross-platform) CUDA RNN, deep autoencoder
Gibson
Apache
Software classification, regression,
Scala,
Foundation, Microsoft Windows, OS Scala, Java, clustering, dimensionality May
MLlib Spark Yes Java, No [215]
UC Berkeley X, Linux Python, R reduction, and collabora- 2014
Python, R
AMPLab, tive filtering
Databricks
Dave C++, Java,
Microsoft Windows, classification, regression, October
Azure Cutler from No C++ ASP.NET, PHP, Yes [216]
Linux clustering 2010
Microsoft Node.js, Python
Ronan
Collobert,
Lua, LuaJIT, C,
Koray Linux, Android, Mac OS October,
Torch Yes C, Lua utility library for Yes deep algorithms [217][218]
Kavukcuoglu, X, iOS 2002
C++/OpenCL
Clement
Farabet
ML algorithms (classifi-
GUI, the cation, regression, cluster-
University of November
MOA Cross-platform Yes Java command-line, No ing, outlier detection, con- [219]
Waikato 2014
and Java cept drift detection and
recommender systems)
Berkeley
Ubuntu, OS X, AWS,
Vision and
unofficial Android port, C++, command
Learning C++, Deep learning algorithms: December
Caffe Windows support by Mi- Yes line, Python, Yes [220]
Center, Python CN, and RNN 2012
crosoft Research, unoffi- MATLAB
community
cial Windows port
contributors
1) Apache Storm: Storm [227] is an open source dis- Foundation on 23 October 2012. Kafka is written in Scala.
tributed real-time data processing framework that provides It supports a variety of use case scenarios with a focus on
massively scalable event collection. The initial release was on high throughput, reliability, and scalability characteristics. For
17 September 2011, it was created by Nathan Marz and the example, it can message sensor data from heating and cooling
team at BackType, and is now owned by Twitter. Storm can equipment in office buildings.
easily process unlimited streams and with any programming
language. It has the capability to process over one million 3) Oracle: In 2013, Oracle started utilizing Oracle Enter-
tuples per second per node with a highly scalable, fault- prise Manager that includes Oracle Big Data Appliance to
tolerant, and reliable architecture. Storm is written in Java and manage all of its big-data technologies. Oracle has also pro-
Clojure. Trident is a high-level abstraction layer for Storm, duced multiple low-latency technologies for Oracle Fast Data
can be utilized to accomplish state management persistence. components includes Oracle Event Processing, Coherence,
Storm is a system of complex event processing. This type of NoSQL, Business Analytics, and Real-Time Decisions. Oracle
solution allows companies to respond to the arrival of sudden Event Processing provides solutions for building applications
and continuous data (information collected in real-time by to filter, correlate and process events in real-time. It supports
sensors, millions of comments generated on social networks IoT services by delivering actionable insight on data streaming
such as Twitter, WhatsApp and Facebook, bank transfers etc.). from a variety of data sources in real-time [229].
Some of the specific applications of Storm include customer Oracle Stream Explorer (OSX) and Oracle R Enterprise
service management in real-time, operational dashboards, data (ORE) aim to support equipment monitoring applications for
monetization, cybersecurity analytics, and threat detection. the systems that made of a variety of components through
sensors, anomaly detection and failure prediction of such sys-
2) Apache Kafka: Kafka [228] is a fast, scalable, fault- tems. ORE [230] is utilized to handle low-frequency streams
tolerant and durable open-source message broker project that in batch mode, while OSX handles the high-frequency streams
originally developed by LinkedIn, and subsequently open making real-time predictions and sends the results back to user
sourced in early 2011 and released by Apache Software applications that are communicating with the output channels.
OSX [231] is a middleware platform has the capability including network traffic data, social-media-generated data,
to process large amounts of streaming data in real-time for email messages and pretty much any data source possible into
a variety of streaming data applications, from a multitude the HDFS.
of sources like sensors, social media, financial feeds, etc. It The architecture of Flume is simple and flexible, it is also
streamlines real-time data delivery into most popular big data robust and fault tolerant with tunable reliability mechanisms
solutions, including Apache Hadoop, Apache HBase, Apache for failover and recovery. log manufacturing operations is an
Hive, Apache Flume, and Apache Kafka to facilitate im- example of Flume’s application. The a massive log file data
proved insight and timely action. Oracle Real-Time Decisions can stream through Flume. The log file data can be stored in
[232] is a decision management platform with self-learning HDFS and analyzed by utilizing Apache Hive.
that determines optimized recommendations and actions with 7) Apache SAMOA: SAMOA [243] is a distributed stream-
messaging, imagery, products, and services within business ing ML framework that contains programming abstractions
processes. for distributed streaming ML algorithms. Its name stands for
4) Amazon Kinesis Streams: Amazon Kinesis [233] is a Scalable Advanced Massive Online Analysis and was origi-
platform for collecting and processing large streams of data nally developed at Yahoo! Labs in Barcelona in 2013 and has
on AWS in real-time, AWS launched Kinesis in November been part of the Apache incubator since late 2014. SAMOA
of 2013, offering powerful services for loading and analyzing is both a platform and a library. It enables the algorithm
streaming data, in addition, it provides custom streaming data developer to reuse their code to run on different underlying
applications for specialized needs. Sometimes Terabytes of execution engine. In addition, it supports plug-in modules
data per hour can be generated that need to be collected, to port SAMOA to different engines. By utilizing SAMOA,
stored, and processed continuously from various application the ML algorithm developer does not need to worry about
services such as web applications, mobile devices, wearables, the complexity of underlying distributed stream processing
industrial sensors etc. Typically, Amazon Kinesis Streams engines. They can run it locally or utilizing one of stream
application can use the Amazon Kinesis Client Library and processing engines, such as Storm, S4, or Samza.
reads data from an Amazon Kinesis stream as data records. SAMOA provides the ML algorithms for a variety of
These applications can run on Amazon EC2 instances. tasks including classification, regression, clustering, along with
boosting, and bagging for ensemble learning. Additionally,
5) Apache Spark Streaming: Apache Spark [234] is an
it offers a platform for the implementation of these ML
open-source platform for real-time data processing, it can
algorithms, as well as a framework that enables the user to
implement using four different languages: Scala, the syntax
write their own distributed streaming algorithms. For example,
in which the platform is written; Python; R; and Java. Spark
there is CluStream for clustering, as well as Vertical Hoeffding
Streaming is an extension of core Spark API. It allows build-
Tree, which uses vertical parallelism on top of the decision
ing fault-tolerant processing of real-time data streams. Spark
tree, or Hoeffding tree for classification. There is also Adaptive
Streaming allows the processing of millions of data among
Model Rules Regressor, which uses both vertical and horizon-
the clusters, and Spark SQL which makes it easier to exploit
tal parallelism implementations for regression [244].
the data through the SQL language. Spark Streaming divides
A summarized comparison between various real-time data
the live data stream into a predefined interval of batches, then
analytics tools is provided in Table VII.
handles each batch of data as Resilient Distributed Datasets
(RDDs). Then we can apply operations like map, reduce, join, V. APPLICATIONS OF ML-B ASED CONTEXT-AWARE
window etc. to process these RDDs. The last results of these S YSTEMS FOR SB S
operations are then returned in batches.
The potential uses of ML in an SB environment can be
Spark Streaming can be utilized for a variety of application divided into four categories: detection, recognition, prediction,
such as real-time monitoring and analyzing of application and optimization [79]. We discuss these categories separately
server logs. These logs messages are considered time se- next.
ries data. Examples of such type of data are sensor data, In general, detection is the extraction of particular infor-
weather information, and clickstream data. This data can also mation from a larger stream of information. Many detection
be utilized for predicting future states based on historical applications in SBs such as fire detection, leak detection,
data. Apache assures a computation speed that performs the and anomaly detection [245]. Many different applications
operations quicker by 100 times than what is currently offered have been studied by researchers in activity recognition in
by Hadoop MapReduce in memory, and 10 times better SBs; examples include fitness tracking, health monitoring, fall
than in disc. Spark can be executed either in independent detection. [246].
cluster mode or in the cloud on different frameworks such The goal of recognition is to classify an object or an event
as Hadoop, Apache Mesos, and EC2. In addition, Spark can to a predefined category. It focuses on how to make computer
access numerous databases such as HDFS, Cassandra, HBase programs perform intelligent and human-like tasks, such as
or S3, Amazon’s data warehouse. the recognition of an object from an image.
6) Apache Flume: Flume [242] is a distributed, reliable The goal of prediction is to determine the temporal re-
and open-source log data aggregation framework. Apache lations’ model between specific events to predict what will
Flume is applied in many applications ranging from log data happen in the near future. Prediction can be either for classi-
aggregation, to transport massive quantities of event data fication or regression problems [247]. Event prediction when
TABLE VII: COMPARISON BETWEEN REAL-TIME DATA ANALYTICS TOOLS
First re- Main Auto- Event Fault Toler-
Tool Platform Written in API languages Type Used in
leased in Owner Scaling? Size ance
Sept. Backtype, Cross- Clojure and Any program- Distributed stream
Storm No Single Yes [235][236]
2011 Twitter platform Java ming language processing
Jan. LinkedIn, Cross- Java, C++, [237][236]
Kafka Scala Yes Single Yes Message broker
2011 Confluent platform Node.js [238]
Java, Node.js,
Jan. Cross- Distributed stream
Oracle Oracle Java Python, PHP, Yes NA Yes [239]
2013 platform processing
and Ruby
Microsoft
May AMPLab, Windows, Scala, Java, Scala, Java, Mini- Streaming
Spark Yes Yes [237][215]
2014 Databricks OS X, Python, R Python, R batch analytics.
Linux
Microsoft Data
C++, Java,
Amazon Dec. Windows, blob of Real-time streaming
AWS C++ Python, Ruby, Yes Yes [240]
Kinesis 2013 OS X, 1 MB data
Node.js, .NET
Linux size
Just with
Jan. Apple, Cross- Distributed stream
Flume Java Java No Single file channel [215]
2012 Cloudera platform processing
only
Created
May Cross- Distributed stream
SAMOA at Yahoo Java Java Yes NA Yes [241]
2013 platform processing
Labs
VI. O PEN I SSUES AND F UTURE R ESEARCH D IRECTIONS technologies and relevant protection laws and tools for identity
Research on SBs has made great strides in recent years, management of users and objects [286]. The recent trend of
but a number of challenges remain. We present some major ML research has focused on handling security and privacy
challenges related to SBs in this part of the work. These issues in SB environments. There are different security-related
challenges will channelize the research directions for future services have utilized ML techniques, such as determining
SBs. safe device behavior by detecting and blocking activities and
potentially harmful behavior [287].
A. Security and Privacy ML techniques have the potential to reduce security gap
Wherever there is an interconnection of two systems or because of their capability to learn, identify and detect the
networks (wired or wireless), there are issues of security and users’ habits and behaviors. Consequently, it can detect the
privacy and the same is true in the case of SB. Security is an abnormal behaviors predicting risks and intrusions before they
essential role in SB environments. Any SB application should happen. For instance, ML models learn the routine of the users,
ensure the confidentiality and integrity of data. Access control such as the time they get home or go to sleep. These models
must be included in SB systems, for instance, the unauthorized can suggest rules based on those detected behaviors from all
users should not be able to disconnect the alarm system by connected devices [288].
connecting the pervasive system [284]. There is a massive
amount of streaming that is collected from the various installed B. SBs and context-aware computing
sensors and appliances, such data needs to be processed and In the SB environment, there exists a massive amount of
stored. Hence, cloud computing services can be utilized for raw data being continuously collected about the various human
this purpose. However, with all of this data that is transmitted, activities and behaviors. It is important to develop techniques
the issue of losing the privacy increases. Therefore, different that convert this raw data into valuable knowledge [289].
encryption techniques are needed to preserve personal privacy Context awareness and ML techniques are expected to provide
[285]. great support to process and store big data and create important
There are specific challenges related to the user’s privacy knowledge from all this data [290].
including challenges related to the data privacy of personal The process of data interpretation and knowledge extrac-
information and the privacy of the individual’s physical lo- tion has the following challenges including addressing noisy
cation and tracking. That needs for privacy enhancement real-world data and the ability to develop further inference
techniques that do not have the limitations of traditional D. Big data challenges in SBs
algorithms. Usually, It is very complex to formalize and model Nowadays, a variety of sensing technology in the SBs can
the contextual information related to human behaviors in a be utilized to collect a massive amount of heterogeneous
standard way due to the complex physiological, psychological data at a reasonable cost. Typically, hundreds of thousands of
and behavioral aspects of human beings [291]. transactions can be generated by a single SB every day. The
The humans communicate through rich languages as well process of storing this data over the long-term is challenging
as gestures and expressions. Modern ubiquitous computer [258]. We can imagine the challenges and opportunities that
systems lack an automatic mechanism of inferring information the companies or government will encounter in the future
as the humans do. New research is necessary to raise human to manage incoming data from dozens of SBs. This new
activities and behaviors recognition to understand the complex data could provide us with more contextual information that
dependencies between the apps and humans [292], [293]. The consequently leads to much better services to the occupants
context-aware prompting systems have essential applications [301].
in SBs such as emergency notifications, medication prompting, In the world of big data, despite the availability massive
heart rate monitoring, generation of agenda reminders, and amount of data, however, it is not necessarily easy to obtain
weather alerts. However, issuing prompts for all detected errors valuable information from this data utilizing the traditional
can possibly be false positives, and consequently, lead to approaches like trial and error to extract meaningful infor-
annoyance and sometimes prove to be unsafe for specific mation from this data. Analyzing these massive amounts of
activities. ML methods can be used for an accurate and precise data requires new technologies to store, organize, and process
prediction when a person faces difficulty while doing daily life big data effectively, it needs high-performance processors that
activities [294]. enable uncovering the insights in big data. It also requires flex-
ible cloud computing services and virtualization techniques,
as well as software such as Apache Hadoop and Spark [302].
C. Personal Data Stream Management in SBs It requires providing appropriate ML techniques which differ
from the traditional approaches for effective and efficient so-
The data streaming management system is able to process
lution of the above issues. For these reasons, researchers have
and transfer raw data collected from a variety of sensors to
recently started to think about the problems and opportunities
information, it is also able to fuse this information to a feature
resulting from the adoption of big data in SB environments
and directly process features [295]. While the data processing
[303] [304]. The information extracted from this big data has
for a single SB is simple, it is more complex when processing
significant value and could greatly contribute in the future of
the data from multiple SBs, because there are different people
SBs as assistive tools and for better services delivery. That is
that tend to share less common interests and have opposing
why it is necessary that the researchers start to analyze and
interests concerning the processed data [296]. The simple
think about the solutions for the current and future challenges
sensors in an SB environment can detect different events
of big data in SBs [305].
related to temperature, motion, light, or weather. Moreover,
other appliances like a television and a telephone can also send
their status or other data as events. All this data from different E. Interoperability
sensors can be used by SB services to detect specific states Interoperability means that two (or more) systems work
and send a request to some actuators according to specific together unchanged even though they were not necessar-
predefined rules, for instance, turn on the light if the television ily designed to work together. When equipment, devices or
is used [297]. appliances having different communication and networking
However, this approach is not generalizable in case of a technologies can communicate effectively, interoperability is
group of people residing in the same building. Although it satisfied. It is a challenge to ensure that an SB that has various
can work well for one certain person when personal prefer- components will be intelligible. Typically, each of these com-
ences can be automatically learned for an individual person, ponents might have been produced by different vendors, each
therefore each of the residents has to define their own set of of which may have created under different design constraints
rules [298]. Because of the increasing number of sensors that and considerations [306]. Therefore it becomes essential to
produce data streams, the traditional analyzing and processing satisfy interoperability so that a number of heterogeneous
techniques of these data streams are mostly impractical now communication and networking technologies could coexist in
[299]. various parts of SBs. For example, an energy management sys-
Despite the availability of new tools and systems for han- tem may use Wi-Fi and ZigBee for communication purposes.
dling massive amounts of data continuously generating by A lot of work can be done in this context [307].
a variety of sensors in SBs, however, the real promise of
advanced data analytics to still lies beyond the realm of pure
technology [296]. In [300] discuses research challenges for F. Reliability
data streams of real-world applications. They analyze issues We can expect that the reliability is one of the main concern
concerning privacy, timing, preprocessing, relational and event of occupants and developers of SB systems. A variety of
streams, model complexity and evaluation, availability of appliances and devices present in SB such as televisions,
information, and problems related to legacy systems. microwave, washing machines etc. are required exceedingly to
be reliable. Achieving expected levels of reliability, especially • Further research is needed into context-aware prompting
when linked with communication technologies utilized with systems, personal data streaming and big data analysis of
these devices that may be expected in SBs, is a great challenge. occupants in SB environment;
There are different reasons for these challenges differences • Some of the other challenges like the interoperability,
in technological approaches, regulations, development culture, reliability, and integration still require more attention.
and the expectations of the market [306].
VII. C ONCLUSIONS
G. Integration
The promise of smart buildings (SBs) is a world of ap-
The key to a successful SB implementation is integration: pliances that anticipate your needs and do exactly what you
linking building systems such as lighting, power meters, want them to at the touch of a button. Since SBs and their
water meters, pumps, heating, and chiller plants together inhabitants create voluminous amounts of streaming data, SB
using sensors and control systems, and then connecting the researchers are looking towards techniques from ML and big
building automation system to enterprise systems. Integration data analytics for managing, processing, and gaining insights
allows executives to gain smart-building benefits, both in new from this big data. This paper reviewed the most important
construction and by gradually transforming existing buildings aspects of SBs with particular focus on what is being done
into SBs. What these SBs have in common is integration. Gen- and what are the issues that require further research in ML
erally, the integration in SB systems brings a range of benefits and data analytics domains. In this regards, we have presented
from energy savings to productivity gains to sustainability. The a comprehensive survey of the research works that relate to
SB systems can be attached to enterprise business systems the use of ML and big data particularly for building smart
to add another level of intelligence that enhances decision- infrastructure and services. Although the recent advancements
making and improves building performance [2]. in technologies that make the concept of SBs feasible, there
However, integrating multiple systems is very challenging are still a variety of challenges that limit large-scale real-world
as each individual system has its own assumptions, strategies systems in SBs field. Addressing these challenges soon will be
to control the physical world, and semantics. As an example a powerful driving force for advancements in both industrial
of integrating two systems in SB, assume a system that and academic fields of SB research.
is responsible for energy management, and another system
for health care are running concurrently. In this case, the
REFERENCES
integrated system should not turn off medical appliances to
save energy while they are being used as suggested by the [1]A. T.-p. So and W. L. Chan, “Applications in Electrical Systems,”
health care system [292]. in Intelligent Building Systems, ser. The International Series on
Asian Studies in Computer and Information Science. Springer,
As a future perspective for SBs, You will wake up to the Boston, MA, 1999, pp. 123–142. [Online]. Available: https:
sound of the alarm, at the same time the available sensors //link.springer.com/chapter/10.1007/978-1-4615-5019-8 13
will be aware that you are waking up. The other sensors such [2]“Improving Performance with Integrated Smart Buildings,” Dec. 2017.
[Online]. Available: http://www.facilitiesnet.com/emails/Building
as light sensors will automatically turn on the light in the Maximizer/Whitepaper/041912.asp
building, while the thermostat will warm the area that you are [3]J. Cheng and T. Kunz, “A Survey on Smart Home Networking,”
about to use in the building. Your coffee will start to brew, you Carleton University, Systems and Computer Engineering,, Tech. Rep.,
Sep. 2009. [Online]. Available: http://kunz-pc.sce.carleton.ca/Thesis/
will also get a notification on your phone about the weather. SmartHomeNetworking.pdf
The other sensors in the kitchen and refrigerator will remind [4]N. Davies, “How the Internet of Things will enable smart buildings,”
you with a list of items that you will need to pick up on your Jul. 2015. [Online]. Available: https://www.extremetech.com/extreme/
209715-how-the-internet-of-things-will-enable-smart-buildings
way from your workplace to home to make dinner. When you
[5]A. Kamilaris, “Enabling smart homes using web technologies,” phdthe-
leave your house, you can press a button from your phone to sis, University of Cyprus, Department of Computer Science, Dec. 2012.
self-drive your car out of the garage. After that, the security [6]“Smart Building Market by Type & region - Global Forecast
system will start monitoring and controlling the home. Such 2022,” MarketsAndMarkets, UK, Technical Report TC 3445,
Aug. 2017. [Online]. Available: https://www.marketsandmarkets.com/
the doors will automatically lock. Appliances will switch to an Market-Reports/smart-building-market-1169.html
energy-saving mode. When the home sensors sense utilizing [7]“Global Forecast for Smart Building Market Size & Share to Surpass
geofencing technology that you are way back home, it will get USD 61,900 million by 2024: Zion Market Research,” Zion Market
Research, Technical Report ZMR-2574, May 2018. [Online]. Available:
ready again for your arrival, the thermostat will warm things http://globenewswire.com/news-release/2018/05/25/1512045/0/en/
up, the garage door will open as you pull up, and your favorite Global-Forecast-for-Smart-Building-Market-Size-Share-to-Surpass-
music will start to play when you walk in [141]. USD-61-900-million-by-2024-Zion-Market-Research.html
[8]“Statista - The Statistics Portal for Market Data, Market Research and
Summary: Although the recent researches have been done Market Studies.” [Online]. Available: https://www.statista.com/
in the SBs field, there is a need for a lot more efforts; however, [9]“Market Insight: The Move From the Connected Home to the
we believe that SBs are possible for the mass market in the Intelligent Home,” Gartner, Inc, Technical Report G00355564,
Jul. 2018. [Online]. Available: https://www.gartner.com/doc/3876868/
near future. The main challenges and future research directions market-insight-connected-home-intelligent
of this eld can be summarized as follows: [10]“IoT in Smart Buildings Market Outlook and Forecasts 2018 -
2023,” ResearchAndMarkets, Global, Technical Report 4539838,
• User context in term of behavior and intention should be May 2018. [Online]. Available: https://www.researchandmarkets.com/
studied and respected whenever possible; reports/4539838/iot-in-smart-buildings-market-outlook-and
[11]“Intelligent CRE for Enterprise: Smart Buildings, Intelligent Science and Engineering. John Wiley & Sons, Inc.,
Workplace, and Management Systems 2018 - 2023,” 2007. [Online]. Available: http://onlinelibrary.wiley.com/doi/10.1002/
Research and Markets, Technical Report 4559384, Jun. 2018. 9780470050118.ecse241/abstract
[Online]. Available: https://www.researchandmarkets.com/reports/ [32]M. V. Moreno, B. beda, A. F. Skarmeta, and M. A. Zamora, “How
4559384/intelligent-cre-for-enterprise-smart-buildings can We Tackle Energy Efficiency in IoT Based Smart Buildings?”
[12]M. Chan, D. Estve, C. Escriba, and E. Campo, “A review of smart Sensors (Basel, Switzerland), vol. 14, no. 6, pp. 9582–9614, May
homesPresent state and future challenges,” Computer Methods and 2014. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc/articles/
Programs in Biomedicine, vol. 91, no. 1, pp. 55–81, Jul. 2008. PMC4118407/
[Online]. Available: http://www.sciencedirect.com/science/article/pii/ [33]O. Hersent, D. Boswarthick, and O. Elloumi, The Internet of Things:
S0169260708000436 Key Applications and Protocols, 2nd ed. Chichester, West Sussex:
[13]M. R. Alam, M. B. I. Reaz, and M. A. M. Ali, “A Review of Smart Wiley, Feb. 2012.
Homes - Past, Present, and Future,” IEEE Transactions on Systems, [34]K. Pohl, G. Bckle, and F. J. v. d. Linden, Software Product Line
Man, and Cybernetics, Part C (Applications and Reviews), vol. 42, Engineering: Foundations, Principles and Techniques, 2005th ed. New
no. 6, pp. 1190–1203, Nov. 2012. York, NY: Springer, Sep. 2005.
[14]G. Lobaccaro, S. Carlucci, E. Lfstrm, G. Lobaccaro, S. Carlucci, and [35]S. C. Mukhopadhyay, “Wearable Sensors for Human Activity Monitor-
E. Lfstrm, “A Review of Systems and Technologies for Smart Homes ing: A Review,” IEEE Sensors Journal, vol. 15, no. 3, pp. 1321–1330,
and Smart Grids,” Energies, vol. 9, no. 5, p. 348, May 2016. [Online]. Mar. 2015.
Available: https://www.mdpi.com/1996-1073/9/5/348 [36]D. Ding, R. A. Cooper, P. F. Pasquina, and L. Fici-Pasquina, “Sensor
[15]J. Pan, R. Jain, and S. Paul, “A Survey of Energy Efficiency in technology for smart homes,” Maturitas, vol. 69, no. 2, pp. 131–136,
Buildings and Microgrids using Networking Technologies,” IEEE Com- Jun. 2011. [Online]. Available: http://www.sciencedirect.com/science/
munications Surveys Tutorials, vol. 16, no. 3, pp. 1709–1731, 2014. article/pii/S0378512211000983
[16]Q. Ni, A. B. Garca Hernando, and I. P. de la Cruz, “The Elderlys [37]D. T.-P. Fong and Y.-Y. Chan, “The use of wearable inertial motion
Independent Living in Smart Homes: A Characterization of Activities sensors in human lower limb biomechanics studies: a systematic
and Sensing Infrastructure Survey to Facilitate Services Development,” review,” Sensors (Basel, Switzerland), vol. 10, no. 12, pp. 11 556–
Sensors, vol. 15, no. 5, pp. 11 312–11 362, May 2015. [Online]. 11 565, 2010.
Available: http://www.mdpi.com/1424-8220/15/5/11312 [38]T. Yilmaz, R. Foster, and Y. Hao, “Detecting vital signs with wearable
[17]P. Rashidi and A. Mihailidis, “A Survey on Ambient-Assisted Living wireless sensors,” Sensors (Basel, Switzerland), vol. 10, no. 12, pp.
Tools for Older Adults,” IEEE Journal of Biomedical and Health 10 837–10 862, 2010.
Informatics, vol. 17, no. 3, pp. 579–590, May 2013. [39]P. Lpez, D. Fernndez, A. J. Jara, and A. F. Skarmeta, “Survey of
[18]K. K. B. Peetoom, M. A. S. Lexis, M. Joore, C. D. Dirksen, and Internet of Things Technologies for Clinical Environments,” in 2013
L. P. De Witte, “Literature review on monitoring technologies and 27th International Conference on Advanced Information Networking
their outcomes in independently living elderly people,” Disability and and Applications Workshops, Mar. 2013, pp. 1349–1354.
Rehabilitation. Assistive Technology, vol. 10, no. 4, pp. 271–294, Jul. [40]V. Smith, T. Sookoor, and K. Whitehouse, “Modeling building thermal
2015. response to HVAC zoning,” ACM SIGBED Review, vol. 9, no. 3, pp. 39–
[19]A. Salih and A. Abraham, “A review of ambient intelligence assisted 45, 2012.
healthcare monitoring,” International Journal of Computer Information [41]N. Forouzandehmehr, S. M. Perlaza, Z. Han, and H. V. Poor, “A
Systems and Industrial Management (IJCISIM), vol. 5, pp. 741–750, satisfaction game for heating, ventilation and air conditioning control
2013. of smart buildings,” Dec. 2013, pp. 3164–3169.
[20]C. Perera, A. Zaslavsky, P. Christen, and D. Georgakopoulos, “Context [42]A. Kottolli, “How Innovations in IoT & Smart Buildings
Aware Computing for The Internet of Things: A Survey,” IEEE are creating Sustainable Work Places,” Dec. 2015. [Online].
Communications Surveys Tutorials, vol. 16, no. 1, pp. 414–454, 2014. Available: https://www.linkedin.com/pulse/how-innovations-iot-smart-
[21]C. W. Tsai, C. F. Lai, M. C. Chiang, and L. T. Yang, “Data Mining buildings-creating-work-places-kottolli/
for Internet of Things: A Survey,” IEEE Communications Surveys
[43]K. Akkaya, I. Guvenc, R. Aygun, N. Pala, and A. Kadri, “IoT-based
Tutorials, vol. 16, no. 1, pp. 77–97, 2014.
occupancy monitoring techniques for energy-efficient smart buildings,”
[22]M. S. Mahdavinejad, M. Rezvan, M. Barekatain, P. Adibi,
in 2015 IEEE Wireless Communications and Networking Conference
P. Barnaghi, and A. P. Sheth, “Machine learning for internet Workshops (WCNCW), Mar. 2015, pp. 58–63.
of things data analysis: a survey,” Digital Communications and Networks,
[44]“WEMO,” Dec. 2017. [Online]. Available: http://www .wemo.com/
vol. 4, no. 3, pp. 161–175, Aug. 2018. [Online]. Available:
[45]“Nest,” Dec. 2017. [Online]. Available: https://www .nest.com/
http://www.sciencedirect.com/science/article/pii/S235286481730247X [23]J.
M. Sinopoli, Smart Buildings Systems for Architects, Owners and Builders, [46]“Lockitron,” Dec. 2017. [Online]. Available: https://lockitron .com/
1st ed. Amsterdam ; Boston: Butterworth-Heinemann, Nov. [47]“SmartThings,” Dec. 2017. [Online]. Available: https://
2009. www.smartthings.com/
[24]“Institute for Building Efficiency,” Sep. 2018. [Online]. Available: [48]“Philips Hue,” Dec. 2017. [Online]. Available: http:
https://buildingefficiencyinitiative.org/ //www2.meethue.com/
[25]F. Zafari, I. Papapanagiotou, and K. Christidis, “Microlocation for [49]“Oleo Apps Inc. , BluFit Bottle,,” Dec. 2017. [Online]. Available:
Internet-of-Things-Equipped Smart Buildings,” IEEE Internet of Things http://www.indigodomo.com/
Journal, vol. 3, no. 1, pp. 96–112, Feb. 2016. [50]“Canary - A complete security system in a single device.” Dec. 2017.
[26]M. Kehoe, M. Cosgrove, S. Gennaro, C. Harrison, W. Harthoorn, [Online]. Available: https://canary.is/
J. Hogan, J. Meegan, P. Nesbitt, and C. Peters, “Smarter cities series: [51]R. Crist and D. Carnoy, “Amazon Echo review,” Dec. 2017. [Online].
a foundation for understanding IBM smarter cities,” Redguides for Available: https://www.cnet.com/products/amazon-echo-review/
Business Leaders, IBM, 2011. [52]“Home - Honeywell Security Group,” Dec. 2017. [Online]. Available:
[27]European Commission and Directorate-General for the Information http://www.security.honeywell.com/index.html
Society and Media, ICT for a low carbon economy: smart electricity [53]K. Karamantas, “Smart home technologies and remote access,” Master
distribution networks. Luxembourg: EUR-OP, 2009. of Science in Networking and Data Communications, Kingston
[28]C. Badica, M. Brezovan, and A. Bdic, “An Overview of Smart University, School Of Computing And Information Systems, UK,
Home Environments: Architectures, Technologies and Applications,” Jul. 2010. [Online]. Available: http://docplayer.net/37654484-Smart-
in CEUR Workshop Proceedings, Thessaloniki, Greece, Sep. 2013, pp. home-technologies-and-remote-access-karamantas-eystratios-master-
78–85. of-science-in-networking-and-data-communications-thesis.html
[29]B. M. Flax, “Intelligent buildings,” IEEE Communications Magazine, [54]C. Perera, C. H. Liu, S. Jayawardena, and M. Chen, “A Survey on
vol. 29, no. 4, pp. 24–27, Apr. 1991. Internet of Things From Industrial Market Perspective,” IEEE Access,
[30]“The Introduction To Smart Home Technologies Information vol. 2, pp. 1660–1679, 2014.
Technology Essay,” Dec. 2017. [Online]. Available: [55]S. Gupta, “Home gateway,” 2004.
https://www.uniassignment.com/essay-samples/information- [56]S. Chernbumroong, A. Atkins, and H. Yu, “Perception of Smart Home
technology/the-introduction-to-smart-home-technologies-information- Technologies to Assist Elderly People,” Paro, Bhutan, 2010.
technology-essay.php [57]P. Carner, “Project Domus: Designing Effective Smart Home Systems,”
[31]R. E. Schantz and D. C. Schmidt, “Middleware for BSc in Information Systems and Information Technology (DT249),
Distributed Systems,” in Wiley Encyclopedia of Computer Dublin Institute of Technology, Dublin, Ireland, Apr. 2009.
[58]“Open Software Platform for the Smart Home,” Dec. e-Health Networking, Applications and Services, Jul. 2010, pp. 322–
2014. [Online]. Available: http://newsroom.cisco.com/press-release- 329.
content?type=webcontent&articleId=1559246 [78]S. Belaidouni and M. Miraoui, “Machine Learning Technologies in
[59]“Mozaiq.” [Online]. Available: https://www .mozaiq.io/ Smart Spaces,” 2016, pp. 52–55.
[60] The Indigo Smart Home System for Mac OS X. [Online]. Available: [79]S. Stenudd, “Using machine learning in the adaptive control of a
http://www.indigodomo.com/ smart environment,” VTT Publications 751, p. 82, 2010. [Online].
[61]“openHAB.” [Online]. Available: http://www .openhab.org/ Available: https://www.vtt.fi/inf/pdf/publications/2010/P751.pdf
[62]L. Smirek, G. Zimmermann, and D. Ziegler, “Towards Universally [80]T. Hastie, R. Tibshirani, and J. Friedman, “Unsupervised Learning,”
Usable Smart Homes How Can MyUI, URC and openHAB Contribute in The Elements of Statistical Learning: Data Mining, Inference, and
to an Adaptive User Interface Platform?” Prediction, Second Edition, 2nd ed. New York, NY: Springer, 2016.
[63]J. Ong, “SmartThings officially launches its connected home platform [81]R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, 2nd ed.
with a new certification program,” May 2014. [Online]. Available: New York: Wiley-Interscience, Nov. 2000.
https://thenextweb.com/insider/2014/05/21/smartthings-officially- [82]J. Lapalu, K. Bouchard, A. Bouzouane, B. Bouchard, and S. Giroux,
launches-connected-home-platform-new-certification-program/ “Unsupervised Mining of Activities for Smart Home Prediction,”
[64]C. Dixon, R. Mahajan, S. Agarwal, A. J. Brush, B. Lee, Procedia Computer Science, vol. 19, pp. 503–510, Jan. 2013.
S. Saroiu, and P. Bahl, “An Operating System for the Home,” [Online]. Available: http://www.sciencedirect.com/science/article/pii/
in Proceedings of the 9th USENIX Conference on Networked S1877050913006753
Systems Design and Implementation, ser. NSDI’12. Berkeley, CA, [83]A. N. Aicha, G. Englebienne, and B. Krse, “Modeling Visit Behaviour
USA: USENIX Association, 2012, pp. 25–25. [Online]. Available: in Smart Homes Using Unsupervised Learning,” in Proceedings
http://dl.acm.org/citation.cfm?id=2228298.2228332 of the 2014 ACM International Joint Conference on Pervasive
[65]F. AL-Riyami, “Microsofts Lab of Things platform is an and Ubiquitous Computing: Adjunct Publication, ser. UbiComp ’14
IoT development playground with life-changing potential.” Adjunct. New York, NY, USA: ACM, 2014, pp. 1193–1200. [Online].
[Online]. Available: https://www.onmsft.com/news/microsofts-lab- Available: http://doi.acm.org/10.1145/2638728.2638809
things-platform-iot-development-playground-life-changing-potential [84]K. Altun, B. Barshan, and O. Tunel, “Comparative study on
[66]A. J. B. Brush, R. Mahajan, and A. Samuel, “Lab of Things: classifying human activities with miniature inertial and magnetic
Simplifying and Scaling Deployments of Experimental Technology sensors,” Pattern Recognition, vol. 43, no. 10, pp. 3605–3620, Oct.
in Homes,” in Handbook of Smart Homes, Health Care and Well- 2010. [Online]. Available: http://linkinghub.elsevier.com/retrieve/pii/
Being, J. van Hoof, G. Demiris, and E. J. Wouters, Eds. Cham: S0031320310001950
Springer International Publishing, 2014. [Online]. Available: [85]M. C. Mozer, “The Neural Network House: AnEnvironmenthat Adapt-
https://doi.org/10.1007/978-3-319-01904-8 35-1 sto its Inhabitants,” 1998, p. 5.
[67]“Lab Of Things,” Jan. 2018. [Online]. Available: https: [86]S. T. M. Bourobou and Y. Yoo, “User Activity Recognition in
//labofthings.codeplex.com/Wikipage?ProjectName=labofthings Smart Homes Using Pattern Clustering Applied to Temporal ANN
[68]G.-M. Jeong, P. H. Truong, T.-Y. Lee, J.-W. Choi, and M. Lee, Algorithm,” Sensors (Basel, Switzerland), vol. 15, no. 5, pp. 11 953–
“Course design for Internet of Things using Lab of Things of Microsoft 11 971, May 2015. [Online]. Available: https://www.ncbi.nlm.nih.gov/
Research,” in 2016 IEEE Frontiers in Education Conference (FIE), Oct. pmc/articles/PMC4481973/
2016, pp. 1–6. [87]S. H. Hsu, M.-H. Wen, H.-C. Lin, C.-C. Lee, and C.-H. Lee, “AIMED-
[69]“Eclipse SmartHome - A Flexible Framework for the Smart Home,” A Personalized TV Recommendation System,” in Interactive TV: a
Jan. 2018. [Online]. Available: https://www.eclipse.org/smarthome/ Shared Experience, ser. Lecture Notes in Computer Science, P. Cesar,
[70]L. Smirek, G. Zimmermann, and M. Beigl, “Just a Smart Home or Your K. Chorianopoulos, and J. F. Jensen, Eds. Springer Berlin Heidelberg,
Smart Home A Framework for Personalized User Interfaces Based 2007, pp. 166–174.
on Eclipse Smart Home and Universal Remote Console,” Procedia [88]A. Fleury, M. Vacher, and N. Noury, “SVM-Based Multimodal Classi-
Computer Science, vol. 98, pp. 107–116, Jan. 2016. [Online]. Available: fication of Activities of Daily Living in Health Smart Homes: Sensors,
http://www.sciencedirect.com/science/article/pii/S1877050916321391 Algorithms, and First Experimental Results,” IEEE Transactions on
[71]J. Eriksson, F. sterlind, N. Finne, N. Tsiftes, A. Dunkels, T. Voigt, Information Technology in Biomedicine, vol. 14, no. 2, pp. 274–283,
R. Sauter, and P. J. Marrn, “COOJA/MSPSim: interoperability testing Mar. 2010.
for wireless sensor networks,” in Proceedings of the 2nd International [89]X. Wu, V. Kumar, J. Ross Quinlan, J. Ghosh, Q. Yang, H. Motoda,
Conference on Simulation Tools and Techniques. ICST (Institute G. J. McLachlan, A. Ng, B. Liu, P. S. Yu, Z.-H. Zhou, M. Steinbach,
for Computer Sciences, Social-Informatics and Telecommunications D. J. Hand, and D. Steinberg, “Top 10 algorithms in data mining,”
Engineering), 2009, p. 27. Knowledge and Information Systems, vol. 14, no. 1, pp. 1–37, Jan.
[72]P. Kugler, P. Nordhus, and B. Eskofier, “Shimmer, Cooja and Contiki: A 2008. [Online]. Available: https://doi.org/10.1007/s10115-007-0114-2
new toolset for the simulation of on-node signal processing algorithms,” [90]Y.-y. SONG and Y. LU, “Decision tree methods: applications
in 2013 IEEE International Conference on Body Sensor Networks, May for classification and prediction,” Shanghai Archives of Psychiatry,
2013, pp. 1–6. vol. 27, no. 2, pp. 130–135, Apr. 2015. [Online]. Available:
[73]M. Bilal, L. O. Oyedele, J. Qadir, K. Munir, S. O. Ajayi, O. O. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4466856/
Akinade, H. A. Owolabi, H. A. Alaka, and M. Pasha, “Big data in [91]L. Rokach and O. Maimon, “Decision Trees,” in Data Mining and
the construction industry: A review of present status, opportunities, Knowledge Discovery Handbook, O. Maimon and L. Rokach, Eds.
and future trends,” Advanced engineering informatics, vol. 30, no. 3, Boston, MA: Springer US, 2005, pp. 165–192. [Online]. Available:
pp. 500–521, 2016. https://doi.org/10.1007/0-387-25465-X 9
[74]M. Bilal, L. O. Oyedele, O. O. Akinade, S. O. Ajayi, H. A. Alaka, H. A. [92]M. Delgado, M. Ros, and M. Amparo Vila, “Correct behavior identifi-
Owolabi, J. Qadir, M. Pasha, and S. A. Bello, “Big data architecture for cation system in a Tagged World,” Expert Systems with Applications,
construction waste analytics (cwa): A conceptual framework,” Journal vol. 36, no. 6, pp. 9899–9906, Aug. 2009. [Online]. Available:
of Building Engineering, vol. 6, pp. 144–156, 2016. http://www.sciencedirect.com/science/article/pii/S095741740900116X
[75]M. Bilal, L. O. Oyedele, J. Qadir, K. Munir, O. O. Akinade, S. O. Ajayi, [93]M. Viswanathan, “Distributed Data Mining in a Ubiquitous Healthcare
H. A. Alaka, and H. A. Owolabi, “Analysis of critical features and Framework,” in Advances in Artificial Intelligence, ser. Lecture Notes
evaluation of bim software: towards a plug-in for construction waste in Computer Science, Z. Kobti and D. Wu, Eds. Springer Berlin
minimization using big data,” International Journal of Sustainable Heidelberg, 2007, pp. 261–271.
Building Technology and Urban Development, vol. 6, no. 4, pp. 211– [94]I. H. Witten, E. Frank, and M. A. Hall, Data Mining: Practical Machine
228, 2015. Learning Tools and Techniques, 3rd ed. Burlington, MA: Morgan
[76]J. Boger, J. Hoey, P. Poupart, C. Boutilier, G. Fernie, and A. Mihailidis, Kaufmann, Jan. 2011.
“A planning system based on Markov decision processes to guide peo- [95]A. Parnandi, K. Le, P. Vaghela, A. Kolli, K. Dantu, S. Poduri,
ple with dementia through activities of daily living,” IEEE Transactions and G. S. Sukhatme, “Coarse In-Building Localization with
on Information Technology in Biomedicine, vol. 10, no. 2, pp. 323–333, Smartphones,” in Mobile Computing, Applications, and Services,
Apr. 2006. ser. Lecture Notes of the Institute for Computer Sciences, Social
[77]A. Fleury, N. Noury, and M. Vacher, “Introducing knowledge in the Informatics and Telecommunications Engineering. Springer, Berlin,
process of supervised classification of activities of Daily Living in Heidelberg, Oct. 2009, pp. 343–354. [Online]. Available: https:
Health Smart Homes,” in The 12th IEEE International Conference on //link.springer.com/chapter/10.1007/978-3-642-12607-9 25
[96]K. Verbert, R. Babuka, and B. De Schutter, “Combining knowledge [112]F. Lv and R. Nevatia, “Recognition and Segmentation of 3-D Human
and historical data for system-level fault diagnosis of HVAC systems,” Action Using HMM and Multi-class AdaBoost,” in Computer Vision
Engineering Applications of Artificial Intelligence, vol. 59, pp. 260– ECCV 2006, ser. Lecture Notes in Computer Science, A. Leonardis,
273, Mar. 2017. [Online]. Available: http://www.sciencedirect.com/ H. Bischof, and A. Pinz, Eds. Springer Berlin Heidelberg, 2006, pp.
science/article/pii/S095219761630255X 359–372.
[97]O. Brdiczka, P. Reignier, and J. L. Crowley, “Detecting Individual Ac- [113]E. Wu, P. Zhang, T. Lu, H. Gu, and N. Gu, “Behavior prediction using
tivities from Video in a Smart Home,” in Knowledge-Based Intelligent an improved Hidden Markov Model to support people with disabilities
Information and Engineering Systems, ser. Lecture Notes in Computer in smart homes,” in 2016 IEEE 20th International Conference on
Science, B. Apolloni, R. J. Howlett, and L. Jain, Eds. Springer Berlin Computer Supported Cooperative Work in Design (CSCWD), May
Heidelberg, 2007, pp. 363–370. 2016, pp. 560–565.
[98]T. v. Kasteren and B. Krose, “Bayesian activity recognition in residence [114]S. Choi, E. Kim, and S. Oh, “Human behavior prediction for smart
for elders,” in 2007 3rd IET International Conference on Intelligent homes using deep learning,” in 2013 IEEE RO-MAN, Aug. 2013, pp.
Environments, Sep. 2007, pp. 209–212. 173–179.
[99]A. Badlani and S. Bhanot, “Smart Home System Design based on [115]F. Ordez, D. Roggen, F. J. Ordez, and D. Roggen, “Deep Convolutional
Artificial Neural Networks,” in Proceedings of the World Congress on and LSTM Recurrent Neural Networks for Multimodal Wearable
Engineering and Computer Science 2011, San Francisco, USA, Oct. Activity Recognition,” Sensors, vol. 16, no. 1, p. 115, Jan. 2016.
2011, pp. 106 –111. [Online]. Available: http://www.mdpi.com/1424-8220/16/1/115
[100]E. Campo, S. Bonhomme, M. Chan, and D. Esteve, “Learning [116]M. A. Alsheikh, A. Selim, D. Niyato, L. Doyle, S. Lin, and H.-P. Tan,
life habits and practices: an issue to the smart home,” in 4th “Deep Activity Recognition Models with Triaxial Accelerometers,”
international conference on smart Homes and health telematic, Oct. 2016, p. 6.
Belfast, Northern Ireland, 2006, pp. 355 – 358. [Online]. Available: [117]M. Baccouche, F. Mamalet, C. Wolf, C. Garcia, and A. Baskurt,
http://ebooks.iospress.nl/volumearticle/574 “Sequential Deep Learning for Human Action Recognition,” in Human
[101]R. Begg and R. Hassan, “Artificial Neural Networks in Smart Homes,” Behavior Understanding, ser. Lecture Notes in Computer Science,
in Designing Smart Homes: The Role of Artificial Intelligence, ser. A. A. Salah and B. Lepri, Eds. Springer Berlin Heidelberg, 2011,
Lecture Notes in Computer Science, J. C. Augusto and C. D. Nugent, pp. 29–39.
Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2006, pp. 146– [118]Y. Chen and Y. Xue, “A Deep Learning Approach to Human Activity
164. [Online]. Available: https://doi.org/10.1007/11788485 9 Recognition Based on Single Accelerometer,” in 2015 IEEE Interna-
[102]M. Ermes, J. Prkk, J. Mntyjrvi, and I. Korhonen, “Detection of Daily tional Conference on Systems, Man, and Cybernetics, Oct. 2015, pp.
Activities and Sports With Wearable Sensors in Controlled and Uncon- 1488–1492.
trolled Conditions,” IEEE Transactions on Information Technology in [119]C. A. Ronao and S.-B. Cho, “Human activity recognition with
Biomedicine, vol. 12, no. 1, pp. 20–26, Jan. 2008. smartphone sensors using deep learning neural networks,” Expert
Systems with Applications, vol. 59, pp. 235–244, Oct. 2016.
[103]L. Ciabattoni, G. Ippoliti, A. Benini, S. Longhi, and M. Pirro,
[Online]. Available: http://www.sciencedirect.com/science/article/pii/
“Design of a Home Energy Management System by Online Neural
S0957417416302056
Networks,” IFAC Proceedings Volumes, vol. 46, no. 11, pp. 677–682,
[120]X. Chen, X. Li, and S. X. Tan, “From robust chip to smart build-
Jan. 2013. [Online]. Available: http://www.sciencedirect.com/science/
ing: CAD algorithms and methodologies for uncertainty analysis of
article/pii/S1474667016330221
building performance,” in 2015 IEEE/ACM International Conference
[104]J. V. Tu, “Advantages and disadvantages of using artificial neural on Computer-Aided Design (ICCAD), Nov. 2015, pp. 457–464.
networks versus logistic regression for predicting medical outcomes,” [121]K. Bouchard, B. Bouchard, and A. Bouzouane, “Regression
Journal of Clinical Epidemiology, vol. 49, no. 11, pp. 1225–1231, analysis for gesture recognition using RFID technology,” in
Nov. 1996. [Online]. Available: http://www.sciencedirect.com/science/ 2014 IEEE International Conference on Pervasive Computing and
article/pii/S0895435696000029 Communication Workshops (PERCOM WORKSHOPS), Mar. 2014,
[105]Y. Fu, Z. Li, H. Zhang, and P. Xu, “Using Support Vector Machine pp. 1–8. [Online]. Available: doi.ieeecomputersociety.org/10.1109/
to Predict Next Day Electricity Load of Public Buildings with Sub- PerComW.2014.6815280
metering Devices,” Procedia Engineering, vol. 121, pp. 1016–1022, [122]A. Jurek, C. Nugent, Y. Bi, and S. Wu, “Clustering-Based Ensemble
Jan. 2015. [Online]. Available: http://www.sciencedirect.com/science/ Learning for Activity Recognition in Smart Homes,” Sensors (Basel,
article/pii/S1877705815029252 Switzerland), vol. 14, no. 7, pp. 12 285–12 304, Jul. 2014. [Online].
[106]Q. C. Nguyen, D. Shin, D. Shin, and J. Kim, “Real-Time Human Available: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4168494/
Tracker Based on Location and Motion Recognition of User for Smart [123]I. Fatima, M. Fahim, Y.-K. Lee, and S. Lee, “Classifier Ensemble
Home,” in 2009 Third International Conference on Multimedia and Optimization for Human Activity Recognition in Smart Homes,”
Ubiquitous Engineering, Jun. 2009, pp. 243–250. in Proceedings of the 7th International Conference on Ubiquitous
[107]J. Petersen, N. Larimer, J. A. Kaye, M. Pavel, and T. L. Hayes, Information Management and Communication, ser. ICUIMC ’13.
“SVM to detect the presence of visitors in a smart home environment,” New York, NY, USA: ACM, 2013, pp. 83:1–83:7. [Online]. Available:
Conference proceedings: ... Annual International Conference of the http://doi.acm.org/10.1145/2448556.2448639
IEEE Engineering in Medicine and Biology Society. IEEE Engineering [124]Y. Guan and T. Pltz, “Ensembles of Deep LSTM Learners for
in Medicine and Biology Society. Annual Conference, vol. 2012, pp. Activity Recognition Using Wearables,” Proc. ACM Interact. Mob.
5850–5853, 2012. Wearable Ubiquitous Technol., vol. 1, no. 2, pp. 11:1–11:28, Jun.
[108]B. Das, D. J. Cook, N. C. Krishnan, and M. Schmitter-Edgecombe, 2017. [Online]. Available: http://doi.acm.org/10.1145/3090076
“One-Class Classification-Based Real-Time Activity Error Detection in [125]N. K. Survadevara, S. C. Mukhopadhyay, and R. K. Rayudu, “Applying
Smart Homes,” IEEE journal of selected topics in signal processing, SARIMA time series to forecast sleeping activity for wellness model
vol. 10, no. 5, pp. 914–923, Aug. 2016. of elderly monitoring in smart home,” in 2012 Sixth International
[109]A. Parnandi, K. Le, P. Vaghela, A. Kolli, K. Dantu, S. Poduri, and Conference on Sensing Technology (ICST), Dec. 2012, pp. 157–162.
G. S. Sukhatme, “Coarse In-Building Localization with Smartphones,” [126]Y. Zhou, Z. Kang, L. Zhang, and C. Spanos, “Causal analysis for non-
in Mobile Computing, Applications, and Services, ser. Lecture Notes of stationary time series in sensor-rich smart buildings,” in 2013 IEEE
the Institute for Computer Sciences, Social Informatics and Telecom- International Conference on Automation Science and Engineering
munications Engineering, T. Phan, R. Montanari, and P. Zerfos, Eds. (CASE), Aug. 2013, pp. 593–598.
Springer Berlin Heidelberg, 2010, pp. 343–354. [127]V. R. Jakkula and D. J. Cook, “Learning temporal relations in smart
[110]B.-C. Cheng, Y.-A. Tsai, G.-T. Liao, and E.-S. Byeon, “HMM machine home data,” in Proceedings of the second International Conference on
learning and inference for Activities of Daily Living recognition,” Technology and Aging. Canada.(To, 2007.
The Journal of Supercomputing, vol. 54, no. 1, pp. 29–42, Oct. 2010. [128]L. G. Fahad, S. F. Tahir, and M. Rajarajan, “Activity Recognition in
[Online]. Available: https://doi.org/10.1007/s11227-009-0335-0 Smart Homes Using Clustering Based Classification,” in 2014 22nd
[111]P. Chahuara, A. Fleury, F. Portet, and M. Vacher, “On-line International Conference on Pattern Recognition, Aug. 2014, pp. 1348–
Human Activity Recognition from Audio and Home Automation 1353.
Sensors: comparison of sequential and non-sequential models in [129]P. Rashidi and D. J. Cook, “Keeping the Resident in the Loop: Adapting
realistic Smart Homes,” Journal of ambient intelligence and smart the Smart Home to the User,” IEEE Transactions on Systems, Man, and
environments, vol. 8, no. 4, pp. 399–422, Jul. 2016. [Online]. Cybernetics - Part A: Systems and Humans, vol. 39, no. 5, pp. 949–959,
Available: https://hal.archives-ouvertes.fr/hal-01336552 Sep. 2009.
[130]P. Rashidi, D. J. Cook, L. B. Holder, and M. Schmitter-Edgecombe, [148]N. Y. Hammerla, S. Halloran, and T. Plotz, “Deep, Convolutional, and
“Discovering Activities to Recognize and Track in a Smart Envi- Recurrent Models for Human Activity Recognition Using Wearables,”
ronment,” IEEE Transactions on Knowledge and Data Engineering, Apr. 2016, pp. 1533 – 1540.
vol. 23, no. 4, pp. 527–539, Apr. 2011. [149]T.-c. Fu, “A review on time series data mining,” Engineering
[131]D. J. Cook, “Learning Setting-Generalized Activity Models for Smart Applications of Artificial Intelligence, vol. 24, no. 1, pp. 164–181,
Spaces,” IEEE intelligent systems, vol. 2010, no. 99, p. 1, Sep. Feb. 2011. [Online]. Available: http://www.sciencedirect.com/science/
2010. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc/articles/ article/pii/S0952197610001727
PMC3068197/ [150]P. Esling and C. Agon, “Time-series Data Mining,” ACM Comput.
[132]T. Liu, H. Yao, R. Ji, Y. Liu, X. Liu, X. Sun, P. Xu, and Z. Zhang, Surv., vol. 45, no. 1, pp. 12:1–12:34, Dec. 2012. [Online]. Available:
“Vision-Based Semi-supervised Homecare with Spatial Constraint,” in http://doi.acm.org/10.1145/2379776.2379788
Advances in Multimedia Information Processing - PCM 2008, ser. [151]F. Chen, P. Deng, J. Wan, D. Zhang, A. V. Vasilakos, and X. Rong,
Lecture Notes in Computer Science, Y.-M. R. Huang, C. Xu, K.-S. “Data Mining for the Internet of Things: Literature Review and
Cheng, J.-F. K. Yang, M. N. S. Swamy, S. Li, and J.-W. Ding, Eds. Challenges,” Int. J. Distrib. Sen. Netw., vol. 2015, pp. 12:12–12:12,
Springer Berlin Heidelberg, 2008, pp. 416–425. Jan. 2015. [Online]. Available: https://doi.org/10.1155/2015/431047
[133]P. N. A. Fahmi, V. Viet, and C. Deok-Jai, “Semi-supervised [152]W. Bhmer and K. Obermayer, “Regression with Linear Factored Func-
fall detection algorithm using fall indicators in smartphone,” in tions,” in Machine Learning and Knowledge Discovery in Databases,
Proceedings of the 6th International Conference on Ubiquitous ser. Lecture Notes in Computer Science, A. Appice, P. P. Rodrigues,
Information Management and Communication - ICUIMC ’12. Kuala V. Santos Costa, C. Soares, J. Gama, and A. Jorge, Eds. Springer
Lumpur, Malaysia: ACM Press, 2012, p. 1. [Online]. Available: International Publishing, 2015, pp. 119–134.
http://dl.acm.org/citation.cfm?doid=2184751.2184890 [153]J. Brownlee, “A Tour of Machine Learning Algorithms,” Nov.
[134]V. Radu, P. Katsikouli, R. Sarkar, and M. K. Marina, “A semi- 2013. [Online]. Available: https://machinelearningmastery.com/a-tour-
supervised learning approach for robust indoor-outdoor detection of-machine-learning-algorithms/
with smartphones,” in Proceedings of the 12th ACM Conference [154]L. Rokach, “Ensemble-based classifiers,” Artificial Intelligence
on Embedded Network Sensor Systems - SenSys ’14. Memphis, Review, vol. 33, no. 1, pp. 1–39, Feb. 2010. [Online]. Available:
Tennessee: ACM Press, 2014, pp. 280–294. [Online]. Available: https://doi.org/10.1007/s10462-009-9124-7
http://dl.acm.org/citation.cfm?doid=2668332.2668347 [155]Y. Bian and J. B. Holland, “Ensemble Learning of QTL Models
[135]D. Guan, W. Yuan, Y. Lee, A. Gavrilov, and S. Lee, “Activity Recogni- Improves Prediction of Complex Traits,” G3 (Bethesda, Md.), vol. 5,
tion Based on Semi-supervised Learning,” in 13th IEEE International no. 10, pp. 2073–2084, Aug. 2015.
Conference on Embedded and Real-Time Computing Systems and [156]M. Usama, J. Qadir, A. Raza, H. Arif, K.-L. A. Yau, Y. Elkhatib,
Applications (RTCSA 2007), Aug. 2007, pp. 469–475. A. Hussain, and A. Al-Fuqaha, “Unsupervised machine learning for
[136]M. C. Mozer, “Lessons from an Adaptive Home,” in Smart Envi- networking: Techniques, applications and research challenges,” arXiv
ronments. Wiley-Blackwell, 2005, pp. 271–294. [Online]. Available: preprint arXiv:1709.06599, 2017.
https://onlinelibrary.wiley.com/doi/abs/10.1002/047168659X.ch12 [157]B. Wooley, Scaling Clustering for the Data Mining Step in Knowledge
[137]D. Li and S. K. Jayaweera, “Reinforcement learning aided smart-home Discovery, 1999. [Online]. Available: http://penguin.lhup.edu/bwooley/
decision-making in an interactive smart grid,” in 2014 IEEE Green presentation SIGCSE99 scalable KD/paper.html
Energy and Systems Conference (IGESC), Nov. 2014, pp. 1–6. [158]L. Rokach and O. Maimon, “Clustering Methods,” in Data Mining and
[138]A. H. Khalili, C. Wu, and H. Aghajan, “Autonomous Learning of Users Knowledge Discovery Handbook, O. Maimon and L. Rokach, Eds.
Preference of Music and Light Services in Smart Home Applications,” Boston, MA: Springer US, 2005, pp. 321–352. [Online]. Available:
in Behavior Monitoring and Interpretation Workshop at German AI https://doi.org/10.1007/0-387-25465-X 15
Conf, Sep. 2009, p. 12. [159]L. Fiorini, F. Cavallo, P. Dario, A. Eavis, and P. Caleb-Solly,
[139]X. Xu, L. Zuo, and Z. Huang, “Reinforcement Learning Algorithms “Unsupervised Machine Learning for Developing Personalised
with Function Approximation: Recent Advances and Applications,” Behaviour Models Using Activity Data,” Sensors (Basel,
Inf. Sci., vol. 261, pp. 1–31, Mar. 2014. [Online]. Available: Switzerland), vol. 17, no. 5, May 2017. [Online]. Available:
https://doi.org/10.1016/j.ins.2013.08.037 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5469639/
[140]A. Desarkar and A. Das, “Big-Data Analytics, Machine Learning [160]P.-N. Tan, M. Steinbach, and V. Kumar, Introduction to Data Mining,
Algorithms and Scalable/Parallel/Distributed Algorithms,” in Internet 1st ed. Boston: Pearson, May 2005.
of Things and Big Data Technologies for Next Generation Healthcare, [161]A. Aztiria, J. C. Augusto, R. Basagoiti, A. Izaguirre, and D. J. Cook,
ser. Studies in Big Data, C. Bhatt, N. Dey, and A. S. Ashour, Eds. “Learning Frequent Behaviors of the Users in Intelligent Environ-
Cham: Springer International Publishing, 2017, pp. 159–197. [Online]. ments,” IEEE Transactions on Systems, Man, and Cybernetics: Systems,
Available: https://doi.org/10.1007/978-3-319-49736-5 8 vol. 43, no. 6, pp. 1265–1278, Nov. 2013.
[141]R. Burbidge and B. Buxton, “An introduction to support vector ma- [162]K. J. Kang, B. Ka, and S. J. Kim, “A Service Scenario Generation
chines for data mining,” in Keynote Papers, Young or12, University Scheme Based on Association Rule Mining for Elderly Surveillance
of Nottingham, Operational Research Society, Operational Research System in a Smart Home Environment,” Eng. Appl. Artif. Intell.,
Society, 2001, pp. 3–15. vol. 25, no. 7, pp. 1355–1364, Oct. 2012. [Online]. Available:
[142]D. Meyer, Support Vector Machines. The Interface to libsvm in http://dx.doi.org/10.1016/j.engappai.2012.02.003
package e1071, Jul. 2018. [163]E. Nazerfard, P. Rashidi, and D. J. Cook, “Using Association Rule
[143]H. Zhao, Q. Hua, H.-B. Chen, Y. Ye, H. Wang, S. X.- Mining to Discover Temporal Relations of Daily Activities,” in Toward
D. Tan, and E. Tlelo-Cuautle, “Thermal-Sensor-Based Occupancy Useful Services for Elderly and People with Disabilities, ser. Lecture
Detection for Smart Buildings Using Machine-Learning Methods,” Notes in Computer Science, B. Abdulrazak, S. Giroux, B. Bouchard,
ACM Transactions on Design Automation of Electronic Systems, H. Pigot, and M. Mokhtari, Eds. Springer Berlin Heidelberg, 2011,
vol. 23, no. 4, pp. 1–21, June 2018. [Online]. Available: http: pp. 49–56.
//dl.acm.org/citation.cfm?doid=3217208.3200904 [164]X. Zhu, “Semi-Supervised Learning Literature Survey,” Computer
[144]X. Chen and X. Lin, “Big Data Deep Learning: Challenges and Science, University of Wisconsin-Madison, Technical Report 1530.
Perspectives,” IEEE Access, vol. 2, pp. 514–525, 2014. [165]S. Sukode, S. Gite, and H. Agrawal, “Context Aware Framework in
[145]M. A. Nielsen, “Neural Networks and Deep Learning,” IoT: A survey,” International Journal, vol. 4, no. 1, 2015.
Determination Press, Dec. 2017. [Online]. Available: http: [166]K. Judah, S. Roy, A. Fern, and T. G. Dietterich, “Reinforcement
//neuralnetworksanddeeplearning.com Learning Via Practice and Critique Advice,” in Proceedings of the
[146] How technology can bridge language gaps, May 2013. [Online]. Twenty-Fourth AAAI Conference on Artificial Intelligence (AAAI-10),
Available: https://www.microsoft.com/en-us/research/blog/technology- 2010, pp. 481 – 486.
can-bridge-language-gaps/ [167]S. Gu, T. Lillicrap, I. Sutskever, and S. Levine, “Continuous
[147]M. Berry, T. E. Potok, P. Balaprakash, H. Hoffmann, and R. Vatsavai, Deep Q-learning with Model-based Acceleration,” in Proceedings
“Machine Learning and Understanding for Intelligent Extreme of the 33rd International Conference on International Conference
Scale Scientific Computing and Discovery,” Advanced Scientific on Machine Learning - Volume 48, ser. ICML’16. New York,
Computing Research (ASCR) Division of the Office of Science, U.S. NY, USA: JMLR.org, 2016, pp. 2829–2838. [Online]. Available:
Department of Energy, Rockville, MD, DOE Workshop Report, Jan. http://dl.acm.org/citation.cfm?id=3045390.3045688
2015. [Online]. Available: https://www.orau.gov/machinelearning2015/ [168]M. Abu-Elkheir, M. Hayajneh, and N. A. Ali, “Data Management
Machine Learning DOE Workshop Report 6.pdf for the Internet of Things: Design Primitives and Solution,” Sensors,
vol. 13, no. 11, pp. 15 582–15 612, Nov. 2013. [Online]. Available: [189]C. O. S. Sorzano, J. Vargas, and A. P. Montano, “A survey
http://www.mdpi.com/1424-8220/13/11/15582 of dimensionality reduction techniques,” arXiv:1403.2877 [cs, q-
[169]A. Karkouch, H. Mousannif, H. Al Moatassime, and T. Noel, “Data bio, stat], Mar. 2014, arXiv: 1403.2877. [Online]. Available:
quality in internet of things: A state-of-the-art survey,” Journal http://arxiv.org/abs/1403.2877
of Network and Computer Applications, vol. 73, pp. 57–81, Sep. [190]Y. Saeys, I. Inza, and P. Larraaga, “A review of feature
2016. [Online]. Available: http://linkinghub.elsevier.com/retrieve/pii/ selection techniques in bioinformatics,” Bioinformatics, vol. 23,
S1084804516301564 no. 19, pp. 2507–2517, Oct. 2007. [Online]. Available: https:
[170]L. Zhao, J. Zhang, and R. Liang, “Data Acquisition and Transmission //academic.oup.com/bioinformatics/article/23/19/2507/185254
System for Building Energy Consumption Monitoring,” Abstract and [191]Y.-L. Chen, X. Wu, T. Li, J. Cheng, Y. Ou, and M. Xu, “Dimensionality
Applied Analysis, vol. 2013, pp. 1–8, 2013. [Online]. Available: reduction of data sequences for human activity recognition,”
http://www.hindawi.com/journals/aaa/2013/613043/ Neurocomputing, vol. 210, pp. 294–302, Oct. 2016. [Online]. Available:
[171]P. Rowley, R. Gough, N. Doylend, A. Thirkill, and P. Leicester, http://www.sciencedirect.com/science/article/pii/S092523121630594X
“From smart homes to smart communities: Advanced data acquisition [192]F. Alas, J. Socor, X. Sevillano, F. Alas, J. C. Socor, and
and analysis for improved sustainability and decision making,” in X. Sevillano, “A Review of Physical and Perceptual Feature
International Conference on Information Society (i-Society 2013), Jun. Extraction Techniques for Speech, Music and Environmental Sounds,”
2013, pp. 263–268. Applied Sciences, vol. 6, no. 5, p. 143, May 2016. [Online]. Available:
[172]“Mmea.” [Online]. Available: http://mmea .fi/ http://www.mdpi.com/2076-3417/6/5/143
[193]O. D. Lara and M. A. Labrador, “A Survey on Human Activity
[173]“e3portaali.” [Online]. Available: http://e3portal .vtt.fi
Recognition using Wearable Sensors,” IEEE Communications Surveys
[174]“List of Home Datasets.” [Online]. Available: http:// Tutorials, vol. 15, no. 3, pp. 1192–1209, 2013.
boxlab.wikispaces.com/List+of+Home+Datasets [194]M. Ramaswami and R. Bhaskaran, “A Study on Feature Selection
[175]A. Yang, A. Giani, and R. Giannatonio, Distributed Human Techniques in Educational Data Mining,” arXiv:0912.3924 [cs],
Action Recognition via Wearable Motion Sensor Networks. [Online]. Dec. 2009, arXiv: 0912.3924. [Online]. Available: http://arxiv.org/abs/
Available: https://people.eecs.berkeley.edu/yang/software/WAR/ 0912.3924
[176]“Activity Recognition in the Home Setting Using Simple [195]M. Hausmann and H. Ziekow, “The Potential of Household Specific
and Ubiquitous Sensors,” Jan. 2018. [Online]. Available: Feature Selection for Analysing Smart Home Time-Series Data,” in
http://courses.media.mit.edu/2004fall/mas622j/04.projects/home/ Smart SysTech 2016; European Conference on Smart Objects, Systems
[177]O. Baos, M. Damas, H. Pomares, I. Rojas, M. A. Tth, and and Technologies, Jul. 2016, pp. 1–6.
O. Amft, “A Benchmark Dataset to Evaluate Sensor Displacement [196]L. G. Fahad, S. F. Tahir, and M. Rajarajan, “Feature selection and
in Activity Recognition,” in Proceedings of the 2012 ACM data balancing for activity recognition in smart homes,” in 2015 IEEE
Conference on Ubiquitous Computing, ser. UbiComp ’12. New International Conference on Communications (ICC), Jun. 2015, pp.
York, NY, USA: ACM, 2012, pp. 1026–1035. [Online]. Available: 512–517.
http://doi.acm.org/10.1145/2370216.2370437 [197]H. Fang, L. He, H. Si, P. Liu, and X. Xie, “Human activity recognition
[178]“CASAS,” Sep. 2018. [Online]. Available: http://casas .wsu.edu/ based on feature selection in smart home using back-propagation
[179]D. Cook, M. Schmitter-Edgecombe, A. Crandall, C. Sanders, and algorithm,” ISA Transactions, vol. 53, no. 5, pp. 1629–1638,
B. Thomas, “Collecting and disseminating smart home sensor data Sep. 2014. [Online]. Available: http://www.sciencedirect.com/science/
in the CASAS project,” in Proceedings of the CHI Workshop on article/pii/S0019057814001281
Developing Shared Home Behavior Datasets to Advance HCI and [198]I. Guyon and A. Elisseeff, “An Introduction to Variable and Feature
Ubiquitous Computing Research, 2009, pp. 1–7. Selection,” J. Mach. Learn. Res., vol. 3, pp. 1157–1182, Mar. 2003.
[180] The PlaceLab. [Online]. Available: http://web.mit.edu/cron/group/ [Online]. Available: http://dl.acm.org/citation.cfm?id=944919.944968
house n/placelab.html [199]H. Fang, R. Srinivasan, and D. J. Cook, “Feature selections for
[181]B. Logan, J. Healey, M. Philipose, E. M. Tapia, and S. Intille, human activity recognition in smart home environments,” International
“A Long-term Evaluation of Sensing Modalities for Activity Journal of Innovative Computing, Information and Control, vol. 8,, pp.
Recognition,” in Proceedings of the 9th International Conference 3525–3535, May 2012.
on Ubiquitous Computing, ser. UbiComp ’07. Berlin, Heidelberg: [200]J. P. Cunningham and Z. Ghahramani, “Linear Dimensionality
Springer-Verlag, 2007, pp. 483–500. [Online]. Available: http: Reduction: Survey, Insights, and Generalizations,” Journal of Machine
//dl.acm.org/citation.cfm?id=1771592.1771620 Learning Research, vol. 16, pp. 2859–2900, 2015. [Online]. Available:
[182]J. Kelly and W. Knottenbelt, “The UK-DALE dataset, domestic http://jmlr.org/papers/v16/cunningham15a.html
appliance-level electricity demand and whole-house demand from five [201]J. Chu, I. Moon, Y. Lee, S. Kim, and M. Mun, “A Supervised Feature-
UK homes,” Scientific Data, vol. 2, p. 150007, Mar. 2015, arXiv: Projection-Based Real-Time EMG Pattern Recognition for
1404.0284. [Online]. Available: http://arxiv.org/abs/1404.0284 Multifunction Myoelectric Hand Control,” IEEE/ASME Transactions
on Mechatronics, vol. 12, no. 3, pp. 282–290, Jun. 2007.
[183]U. Sivarajah, M. M. Kamal, Z. Irani, and V. Weerakkody,
[202]S. Landset, T. M. Khoshgoftaar, A. N. Richter, and T. Hasanin, “A
“Critical analysis of Big Data challenges and analytical methods,”
survey of open source tools for machine learning with big data in the
Journal of Business Research, vol. 70, pp. 263–286, Jan. 2017.
Hadoop ecosystem,” Journal of Big Data, vol. 2, no. 1, p. 24, Nov.
[Online]. Available: http://www.sciencedirect.com/science/article/pii/
2015. [Online]. Available: https://doi.org/10.1186/s40537-015-0032-1
S014829631630488X
[203] H2O platform. [Online]. Available: https://www.h2o.ai/
[184]C. C. Aggarwal, N. Ashish, and A. Sheth, “The Internet of [204]“MLlib | Apache Spark.” [Online]. Available: http://spark.apache.org/
Things: A Survey from the Data-Centric Perspective,” in Managing mllib/
and Mining Sensor Data, C. C. Aggarwal, Ed. Boston, MA: [205]“TensorFlow.” [Online]. Available: https://www .tensorflow.org/
Springer US, 2013, pp. 383–428. [Online]. Available: https: [206]“Torch | Scientific computing for LuaJIT.” [Online]. Available:
//doi.org/10.1007/978-1-4614-6309-2 12 http://torch.ch/
[185]C. C. Aggarwal and J. Han, “A Survey of RFID Data Processing,” [207]J. Murphy, “Deep Learning Frameworks: A Survey of TensorFlow,
in Managing and Mining Sensor Data, C. C. Aggarwal, Ed. Torch, Theano, Caffe, Neon, and the IBM Machine Learning Stack,”
Boston, MA: Springer US, 2013, pp. 349–382. [Online]. Available: Jan. 2016. [Online]. Available: https://www.microway.com/hpc-tech-
https://doi.org/10.1007/978-1-4614-6309-2 11 tips/deep-learning-frameworks-survey-tensorflow-torch-theano-caffe-
[186]S. Gill, B. Lee, and E. Neto, “Context aware model-based cleaning neon-ibm-machine-learning-stack/
of data streams,” in 2015 26th Irish Signals and Systems Conference [208]M.-H. Park, Y.-H. Jang, Y.-W. Ju, and S.-C. Park, “Design of
(ISSC), Jun. 2015, pp. 1–6. TensorFlow-Based Proactive Smart Home Managers,” in Advances in
[187]K. Kenda, J. krbec, and M. krjanc, “Usage Of The Kalman Filter Computer Science and Ubiquitous Computing, ser. Lecture Notes in
For Data Cleaning Of Sensor Data,” in Proceedings of the 16th Electrical Engineering, J. J. Park, V. Loia, G. Yi, and Y. Sung, Eds.
International Multiconference Information Society, Ljubljana, 2013, pp. Springer Singapore, 2018, pp. 83–89.
172–175. [209]J. Vzquez-Canteli, S. Ulyanin, J. Kmpf, and Z. Nagy, “Fusing
[188]A. Ghodsi, “Dimensionality Reduction A Short Tutorial,” Department TensorFlow with building energy simulation for intelligent energy
of Statistics and Actuarial Science, Univ. of Waterloo, Ontario, Canada, management in smart cities,” Sustainable Cities and Society,
vol. 37, 2006. [Online]. Available: https://www.math.uwaterloo.ca/ Nov. 2018. [Online]. Available: http://www.sciencedirect.com/science/
∼aghodsib/courses/f06stat890/readings/tutorial stat890.pdf article/pii/S2210670718314380
[210]H.-Y. Lin, Y.-L. Hsueh, and W.-N. Lie, “Abnormal Event Detection [230]“Oracle Stream Explorer Platform | SOA | Oracle.” [Online].
Using Microsoft Kinect in a Smart Home,” in 2016 International Available: http://www.oracle.com/us/products/middleware/soa/stream-
Computer Symposium (ICS). Chiayi, Taiwan: IEEE, Dec. 2016, explorer-platform/overview/index.html
pp. 285–289. [Online]. Available: http://ieeexplore.ieee.org/document/ [231] Oracle R Enterprise. [Online]. Available: https:
7858486/ //www.oracle.com/technetwork/database/database-technologies/r/r-
[211]W. Wang and C. Miao, “Activity Recognition in New Smart Home enterprise/overview/index.html
Environments,” in Proceedings of the 3rd International Workshop on [232]“Oracle Real-Time Decisions | Oracle.” [Online]. Avail-
Multimedia for Personal Health and Health Care - HealthMedia’18. able: https://www.oracle.com/solutions/business-analytics/business-
Seoul, Republic of Korea: ACM Press, 2018, pp. 29–37. [Online]. intelligence/real-time-decisions/
Available: http://dl.acm.org/citation.cfm?doid=3264996.3265001 [233]“Amazon Kinesis.” [Online]. Available: https://aws .amazon.com/
[212]B. Qolomany, A. Al-Fuqaha, D. Benhaddou, and A. Gupta, “Role kinesis/
of Deep LSTM Neural Networks and Wi-Fi Networks in Support [234]“Spark Streaming | Apache Spark.” [Online]. Available: http:
of Occupancy Prediction in Smart Buildings,” in 2017 IEEE 19th //spark.apache.org/streaming/
International Conference on High Performance Computing and Com- [235]P. Wang, F. Ye, and X. Chen, “A Smart Home Gateway
munications; IEEE 15th International Conference on Smart City; Platform for Data Collection and Awareness,” arXiv:1804.01242
IEEE 3rd International Conference on Data Science and Systems [cs], Apr. 2018, arXiv: 1804.01242. [Online]. Available: http:
(HPCC/SmartCity/DSS), Dec. 2017, pp. 50–57. //arxiv.org/abs/1804.01242
[213]B. Qolomany, M. Maabreh, A. Al-Fuqaha, A. Gupta, and D. Benhad- [236]M. Syafrudin, N. L. Fitriyani, D. Li, G. Alfian, J. Rhee,
dou, “Parameters optimization of deep learning models using Particle and Y.-S. Kang, “An Open Source-Based Real-Time Data
swarm optimization,” in 2017 13th International Wireless Communi- Processing Architecture Framework for Manufacturing Sustainability,”
cations and Mobile Computing Conference (IWCMC), Jun. 2017, pp. Sustainability, vol. 9, no. 11, p. 2139, Nov. 2017. [Online]. Available:
1285–1290. https://www.mdpi.com/2071-1050/9/11/2139
[214]J. A. C. Soto, M. Jentsch, D. Preuveneers, and E. Ilie-Zudor, “CEML: [237]J. Lohokare, R. Dani, A. Rajurkar, and A. Apte, “An IoT ecosystem for
Mixing and moving complex event processing and machine learning the implementation of scalable wireless home automation systems at
to the edge of the network for IoT applications,” in Proceedings smart city level,” in TENCON 2017 - 2017 IEEE Region 10 Conference,
of the 6th International Conference on the Internet of Things - Nov. 2017, pp. 1503–1508.
IoT’16. Stuttgart, Germany: ACM Press, 2016, pp. 103–110. [Online]. [238]A. Petheriotis, “IoT for Smart Homes and trillions of messages
Available: http://dl.acm.org/citation.cfm?doid=2991561.2991575 from Kafka to Elasticsearch,” Mar. 2018. [Online]. Available: https:
[215]M. R. Bashir and A. Q. Gill, “Towards an IoT Big Data //www.landoop.com/blog/2018/03/kafka-elasticsearch-kubernetes-iot/
Analytics Framework: Smart Buildings Systems,” in 2016 IEEE
[239]X. Liu and P. S. Nielsen, “Streamlining Smart Meter Data Analytics,”
18th International Conference on High Performance Computing and
in Proceedings of the 10th Conference on Sustainable Development
Communications; IEEE 14th International Conference on Smart City;
of Energy, Water and Environment Systems. Technical University of
IEEE 2nd International Conference on Data Science and Systems
Denmark, Denmark: DTU library, 2015, p. 15.
(HPCC/SmartCity/DSS). Sydney, Australia: IEEE, Dec. 2016, pp.
[240]“Supercell Case Study Amazon Web Services (AWS).” [Online].
1325–1332. [Online]. Available: http://ieeexplore.ieee.org/document/
Available: https://aws.amazon.com/solutions/case-studies/supercell/
7828529/
[216]B. Van Hoof, “View chooses Microsoft Azure IoT to accelerate devel- [241]H. Elkhoukhi, Y. NaitMalek, A. Berouine, M. Bakhouya,
opment of smart building applications,” Nov. 2018. [Online]. Avail- D. Elouadghiri, and M. Essaaidi, “Towards a Real-time Occupancy
able: https://azure.microsoft.com/en-us/blog/view-chooses-microsoft- Detection Approach for Smart Buildings,” Procedia Computer
azure-iot-to-accelerate-development-of-smart-building-applications/ Science, vol. 134, pp. 114–120, 2018. [Online]. Available:
[217]N. D. Lane, S. Bhattacharya, P. Georgiev, C. Forlivesi, and F. Kawsar, https://linkinghub.elsevier.com/retrieve/pii/S1877050918311141
“An Early Resource Characterization of Deep Learning on Wearables, [242] Apache Flume. [Online]. Available: https://flume.apache.org/
Smartphones and Internet-of-Things Devices,” in Proceedings of [243]“Apache SAMOA.” [Online]. Available: https://
the 2015 International Workshop on Internet of Things towards samoa.incubator.apache.org/
Applications - IoT-App ’15. Seoul, South Korea: ACM Press, 2015, [244]G. De Francisci Morales, “SAMOA: a platform for mining big
pp. 7–12. [Online]. Available: http://dl.acm.org/citation.cfm?doid= data streams,” in Proceedings of the 22nd International Conference
2820975.2820980 on World Wide Web - WWW ’13 Companion. Rio de Janeiro,
[218]“Torch AI: Unleashing the Power of Network- Brazil: ACM Press, 2013, pp. 777–778. [Online]. Available:
Centric Intelligence.” [Online]. Available: https: http://dl.acm.org/citation.cfm?doid=2487788.2488042
//smart-solutions.govciooutlook.com/vendor/torch-ai-unleashing- [245]S.-S. Choi, M.-J. Chun, Y.-S. Lee, and H.-R. Lee, “A Practical Method-
the-power-of-networkcentric-intelligence-cid-133-mid-18.html ology and Framework for Comprehensive Incident Handling Focused
[219]“Massive Online Analysis (MOA),” Jan. 2011. [Online]. Available: on Bot Response,” in Future Generation Information Technology, ser.
https://moa.cms.waikato.ac.nz/related-projects/ Lecture Notes in Computer Science, T.-h. Kim, Y.-h. Lee, B.-H. Kang,
[220]G. Mittal, K. B. Yagnik, M. Garg, and N. C. Krishnan, “SpotGarbage: and D. lzak, Eds. Springer Berlin Heidelberg, 2010, pp. 481–492.
smartphone app to detect garbage using deep learning,” in [246]J. W. Lockhart, T. Pulickal, and G. M. Weiss, “Applications of
Proceedings of the 2016 ACM International Joint Conference on Mobile Activity Recognition,” in Proceedings of the 2012 ACM
Pervasive and Ubiquitous Computing - UbiComp ’16. Heidelberg, Conference on Ubiquitous Computing, ser. UbiComp ’12. New
Germany: ACM Press, 2016, pp. 940–945. [Online]. Available: York, NY, USA: ACM, 2012, pp. 1054–1058. [Online]. Available:
http://dl.acm.org/citation.cfm?doid=2971648.2971731 http://doi.acm.org/10.1145/2370216.2370441
[221]“Deeplearning4j.” [Online]. Available: https://deeplearning4j .org/ [247]C. Nyce, Predictive Analytics White Paper. American Institute
[222]“MOA.” [Online]. Available: https://moa .cms.waikato.ac.nz/ for Chartered Property Casualty Underwriters/Insurance Institute of
[223]A. Bifet, G. Holmes, R. Kirkby, and B. Pfahringer, “MOA: America,, 2007.
Massive Online Analysis,” Journal of Machine Learning Research, [248]S. Stenudd, “A Model for Using Machine Learning in Smart Envi-
vol. 11, no. May, pp. 1601–1604, 2010. [Online]. Available: ronments,” in Grid and Pervasive Computing Workshops, ser. Lecture
http://www.jmlr.org/papers/v11/bifet10a.html Notes in Computer Science, M. Rautiainen, T. Korhonen, E. Muta-
[224]“Caffe | Deep Learning Framework.” [Online]. Available: http: fungwa, E. Ovaska, A. Katasonov, A. Evesti, H. Ailisto, A. Quigley,
//caffe.berkeleyvision.org/ J. Hkkil, N. Milic-Frayling, and J. Riekki, Eds. Springer Berlin
[225]Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, Heidelberg, 2012, pp. 24–33.
S. Guadarrama, and T. Darrell, “Caffe: Convolutional Architecture for [249]S. Abrahms, “Technology, Gadgets For Seniors Aging In Place,” Mar.
Fast Feature Embedding,” arXiv:1408.5093 [cs], Jun. 2014, arXiv: 2014. [Online]. Available: http://www.aarp.org/home-family/personal-
1408.5093. [Online]. Available: http://arxiv.org/abs/1408.5093 technology/info-2014/is-this-the-end-of-the-nursing-home.html
[226] Microsoft Azure Machine Learning. [Online]. Available: https: [250]R. Das, “Design and Implementation of a Smart Home for the El-
//azure.microsoft.com/en-us/services/machine-learning-studio/ derly and Disabled,” International Journal of Computer Networks and
[227]“Apache Storm.” [Online]. Available: http://storm .apache.org/ Applications, vol. 2, no. 6, pp. 242 – 246, Dec. 2015.
[228]“Apache Kafka.” [Online]. Available: http://kafka .apache.org/ [251]T. Taleb, D. Bottazzi, M. Guizani, and H. Nait-Charif, “Angelah: a
[229] Oracle Fusion Middleware. [Online]. Available: https:// framework for assisting elders at home,” IEEE Journal on Selected
docs.oracle.com/middleware/1213/eventprocessing/ Areas in Communications, vol. 27, no. 4, pp. 480–494, May 2009.
[252]M. N. K. Boulos, A. Rocha, A. Martins, M. E. Vicente, [269]“Transit Security Design Considerations Final Report November 2004,”
A. Bolz, R. Feld, I. Tchoudovski, M. Braecklein, J. Nelson, Final Report, p. 341, 2004.
G. Laighin, C. Sdogati, F. Cesaroni, M. Antomarini, A. Jobes, [270]A. Kaklauskas and R. Gudauskas, “17 - Intelligent decision-support
and M. Kinirons, “CAALYX: a new generation of location- systems and the Internet of Things for the smart built environment,” in
based services in healthcare,” International Journal of Health Start-Up Creation, F. Pacheco-Torgal, E. Rasmussen, C.-G. Granqvist,
Geographics, vol. 6, p. 9, Mar. 2007. [Online]. Available: https: V. Ivanov, A. Kaklauskas, and S. Makonin, Eds. Woodhead
//www.ncbi.nlm.nih.gov/pmc/articles/PMC1828720/ Publishing, Jan. 2016, pp. 413–449. [Online]. Available: http:
[253]M. N. Kamel Boulos, R. C. Lou, A. Anastasiou, C. D. Nugent, //www.sciencedirect.com/science/article/pii/B9780081005460000170
J. Alexandersson, G. Zimmermann, U. Cortes, and R. Casas, [271]Q. Ding, Z. Peng, T. Liu, and Q. Tong, “Multi-Sensor Building Fire
“Connectivity for Healthcare and Well-Being Management: Examples Alarm System with Information Fusion Technology Based on D-S
from Six European Projects,” International Journal of Environmental Evidence Theory,” Algorithms, vol. 7, no. 4, pp. 523–537, Oct. 2014.
Research and Public Health, vol. 6, no. 7, pp. 1947–1971, Jul. [Online]. Available: http://www.mdpi.com/1999-4893/7/4/523
2009. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc/articles/ [272] Clarifai. [Online]. Available: https://www.clarifai.com/
PMC2738891/ [273]J. Bangali and A. Shaligram, “Design and Implementation of Security
[254]M. S. Hossain, M. A. Rahman, and G. Muhammad, “Cyberphysical Systems for Smart Home based on GSM technology,” International
cloud-oriented multi-sensory smart home framework for elderly people: Journal of Smart Home, vol. 7, no. 6, pp. 201–208, Nov. 2013. [Online].
An energy efficiency perspective,” Journal of Parallel and Distributed Available: http://www.sersc.org/journals/IJSH/vol7 no6 2013/19.pdf
Computing, vol. 103, pp. 11–21, May 2017. [Online]. Available: [274]J.-V. Lee, Y.-D. Chuah, and C.-T. Chai, “A Multilevel Home
http://www.sciencedirect.com/science/article/pii/S074373151630123X Security System (MHSS),” International Journal of Smart Home,
[255]N. K. Suryadevara, A. Gaddam, R. K. Rayudu, and S. C. vol. 7, no. 2, pp. 49–60. [Online]. Available: http://www.earticle.net/
Mukhopadhyay, “Wireless sensors network based safe home to care article.aspx?sn=204606
elderly people: Behaviour detection,” Sensors and Actuators A: [275]Y. Zhao and Z. Ye, “A low cost GSM/GPRS based wireless home
Physical, vol. 186, pp. 277–283, Oct. 2012. [Online]. Available: security system,” IEEE Transactions on Consumer Electronics, vol. 54,
http://www.sciencedirect.com/science/article/pii/S0924424712001847 no. 2, pp. 567–572, May 2008.
[256]Z. J. Yu, F. Haghighat, and B. C. M. Fung, “Advances and challenges [276] Aware Home Research Initiative. [Online]. Available: http:
in building engineering and data mining applications for energy- //www.awarehome.gatech.edu/
efficient communities,” Sustainable Cities and Society, vol. 25, pp. 33– [277]“House n The PlaceLab,” Sep. 2018. [Online]. Available: http:
38, Aug. 2016. [Online]. Available: http://www.sciencedirect.com/ //web.mit.edu/cron/group/house n/placelab.html
science/article/pii/S2210670715001067
[278]B. Brumitt, B. Meyers, J. Krumm, A. Kern, and S. Shafer, “EasyLiving:
[257]J. King and C. Perry, “Smart Buildings: Using Smart Technology to Technologies for Intelligent Environments,” in Handheld and Ubiqui-
Save Energy in Existing Buildings,” American Council for an Energy- tous Computing, ser. Lecture Notes in Computer Science, P. Thomas
Efficient Economy, Washington, DC, USA, Technical Report A1701, and H.-W. Gellersen, Eds. Springer Berlin Heidelberg, 2000, pp. 12–
Feb. 2017. [Online]. Available: https://aceee.org/research-report/a1701 29.
[258]M. Rodrguez Fernndez, A. Corts Garca, I. Gonzlez Alonso, and
[279]“Mobile & Pervasive Computing at the University of Florida.”
E. Zalama Casanova, “Using the Big Data generated by the
[Online]. Available: https://www.cise.ufl.edu/∼helal/gt.htm
Smart Home to improve energy efficiency management,” Energy
[280]S. Helal, W. Mann, H. El-Zabadani, J. King, Y. Kaddoura, and
Efficiency, vol. 9, no. 1, pp. 249–260, Jan. 2016. [Online]. Available:
https://doi.org/10.1007/s12053-015-9361-3 E. Jansen, “The Gator Tech Smart House: a programmable pervasive
space,” Computer, vol. 38, no. 3, pp. 50–60, Mar. 2005.
[259]C. Reinisch, M. Kofler, F. Iglesias, and W. Kastner, “ThinkHome
Energy Efficiency in Future Smart Homes,” EURASIP Journal on [281]K. Bouchard, S. Giroux, R. Radziszewski, M. Gagnon, Q. Szymanski,
Embedded Systems, vol. 2011, no. 1, p. 104617, Sep. 2010. [Online]. S. Pinard, M. Levasseur, and N. Bier, “Maintaining Good Relationships
Available: https://doi.org/10.1155/2011/104617 in Clinical Setting: The Bonus DOMUS Project,” in Proceedings of
[260]M. Jahn, M. Jentsch, C. R. Prause, F. Pramudianto, A. Al-Akkad, and the 8th ACM International Conference on PErvasive Technologies
R. Reiners, “The Energy Aware Smart Home,” in 2010 5th Interna- Related to Assistive Environments, ser. PETRA ’15. New York,
tional Conference on Future Information Technology, May 2010, pp. NY, USA: ACM, 2015, pp. 8:1–8:4. [Online]. Available: http:
1–8. //doi.acm.org/10.1145/2769493.2769554
[261] Hydra middleware, Mar. 2016. [Online]. Available: https: [282]T. Tamura, A. Kawarada, M. Nambu, A. Tsukada, K. Sasaki, and K.-I.
//vicinity2020.eu/vicinity/content/hydra Yamakoshi, “E-healthcare at an experimental welfare techno house in
[262]J. Pan, R. Jain, S. Paul, T. Vu, A. Saifullah, and M. Sha, “An Internet of Japan,” The Open Medical Informatics Journal, vol. 1, pp. 1–7, 2007.
Things Framework for Smart Energy in Buildings: Designs, Prototype, [283]T. Yamazaki, “The Ubiquitous Home,” International Journal of Smart
and Experiments,” IEEE Internet of Things Journal, vol. 2, no. 6, pp. Home, vol. 1, no. 1, pp. 17–22, 2007.
527–537, Dec. 2015. [284]M. A. Al-Qutayri and J. S. Jeedella, “Smart homes: technologies,
[263]A. Fensel, S. Tomic, V. Kumar, M. Stefanovic, S. V. Aleshin, and D. O. applications and challenges,” International Journal of Computer Aided
Novikov, “SESAME-S: Semantic Smart Home System for Energy Engineering and Technology, vol. 2, no. 2-3, pp. 125–144, Dec.
Efficiency,” Informatik-Spektrum, vol. 36, no. 1, pp. 46–57, Feb. 2013. 2009. [Online]. Available: https://www.inderscienceonline.com/doi/
[Online]. Available: https://doi.org/10.1007/s00287-012-0665-9 abs/10.1504/IJCAET.2010.030541
[264]N. Vastardis, M. Kampouridis, and K. Yang, “A user behaviour-driven [285]K. Islam, W. Shen, and X. Wang, “Security and privacy considerations
smart-home gateway for energy management,” Journal of Ambient for Wireless Sensor Networks in smart home environments,” in Pro-
Intelligence and Smart Environments, vol. 8, no. 6, pp. 583–602, ceedings of the 2012 IEEE 16th International Conference on Computer
Jan. 2016. [Online]. Available: https://content.iospress.com/articles/ Supported Cooperative Work in Design (CSCWD), May 2012, pp. 626–
journal-of-ambient-intelligence-and-smart-environments/ais403 633.
[265]O. Vermesan, Internet of things applications - from research and [286]D. Bandyopadhyay and J. Sen, “Internet of Things: Applications and
innovation to market deployment. Place of publication not identified: Challenges in Technology and Standardization,” Wireless Personal
River Publishers, 2014, oCLC: 884398675. Communications, vol. 58, no. 1, pp. 49–69, May 2011. [Online].
[266]S. S. Khan and J. Hoey, “Review of fall detection techniques: A data Available: https://doi.org/10.1007/s11277-011-0288-5
availability perspective,” Medical Engineering & Physics, vol. 39, pp. [287]P. Smitka, Get Ready for the Intelligent Workplace, Jun. 2016. [Online].
12–22, Jan. 2017. [Online]. Available: http://www.sciencedirect.com/ Available: https://peter.smitka.net/data/Intelligent Workplace v1.1.pdf
science/article/pii/S1350453316302600 [288]K. Rose, S. Eldridge, and L. Chapin, “The in-
[267]M. H. Kabir, M. R. Hoque, and S.-H. Yang, “Development of a ternet of things: An overview,” Oct. 2015. [Online].
Smart Home Context-aware Application: A Machine Learning based Available: https://www.internetsociety.org/wp-content/uploads/2017/
Approach,” International Journal of Smart Home, vol. 9, no. 1, 08/ISOC-IoT-Overview-20151221-en.pdf
pp. 217–226, Jan. 2015. [Online]. Available: http://www.sersc.org/ [289]D. Zhang, B. Guo, and Z. Yu, “The Emergence of Social and Com-
journals/IJSH/vol9 no1 2015/23.pdf munity Intelligence,” Computer, vol. 44, no. 7, pp. 21–28, Jul. 2011. [290]A.
[268]J. Ahn, D. Shin, K. Kim, and J. Yang, “Indoor Air Quality Ali, J. Qadir, R. u. Rasool, A. Sathiaseelan, A. Zwitter, and
Analysis Using Deep Learning with Sensor Data,” Sensors (Basel, J. Crowcroft, “Big data for development: applications and techniques,”
Switzerland), vol. 17, no. 11, Oct. 2017. [Online]. Available: Big Data Analytics, vol. 1, no. 1, p. 2, Jul. 2016. [Online]. Available:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5712838/ https://doi.org/10.1186/s41044-016-0002-4
[291]B. C. Ooi, K. L. Tan, Q. T. Tran, J. W. Yip, G. Chen, Basheer Qolomany (S’17) received the Ph.D. and
Z. J. Ling, T. Nguyen, A. K. Tung, and M. Zhang, “Contextual second masters en-route to Ph.D. degrees in Com-
Crowd Intelligence,” SIGKDD Explor. Newsl., vol. 16, no. 1, pp. 39– puter Science from Western Michigan University
46, Sep. 2014. [Online]. Available: http://doi.acm.org/10.1145/ (WMU), Kalamazoo, MI, USA, in 2018. He also
2674026.2674032 received his B.Sc. and M.Sc. degrees in computer
[292]R. Alur, E. Berger, A. W. Drobnis, L. Fix, K. Fu, G. D. Hager, science from University of Mosul, Mosul, Iraq,
D. Lopresti, K. Nahrstedt, E. Mynatt, S. Patel, J. Rexford, J. A. in 2008 and 2011, respectively. He is currently
Stankovic, and B. Zorn, “Systems Computing Challenges in the Internet an Assistant Professor at Department of Computer
of Things,” p. 15. Science, Kennesaw State University, Marietta, GA,
[293]J. A. Stankovic, “Research Directions for the Internet of Things,” IEEE USA. Previously, he served as a Graduate Doc-
Internet of Things Journal, vol. 1, no. 1, pp. 3–9, Feb. 2014. toral Assistant at Department of Computer Science,
[294]W. K. Edwards and R. E. Grinter, “At Home with Ubiquitous Com- WMU, in 2016-2018; he also served as a lecturer at Department of Computer
puting: Seven Challenges,” in Ubicomp 2001: Ubiquitous Computing, Science, University of Duhok, Kurdistan region of Iraq, Iraq, in 2011-2013.
ser. Lecture Notes in Computer Science, G. D. Abowd, B. Brumitt, His research interests include machine learning, deep learning, Internet of
and S. Shafer, Eds. Springer Berlin Heidelberg, 2001, pp. 256–272. Things, smart services, cloud computing, and big data analytics.
[295]M. Stonebraker, U. etintemel, and S. Zdonik, “The 8 requirements Dr. Qolomany has served as a Technical Program Committee (TPC)
of real-time stream processing,” ACM SIGMOD Record, vol. 34, member and a reviewer of some international conferences include: IWCMC
no. 4, pp. 42–47, Dec. 2005. [Online]. Available: http://portal.acm.org/ 2018, VTC 2018, MEDES 2016, and IC4 2016.
citation.cfm?doid=1107499.1107504
[296]D. Geesen, H. J. Appelrath, M. Grawunder, and D. Nicklas,
“Challenges for Personal Data Stream Management in
Smart Buildings,” Creating Personal, Social, and Urban
Awareness through Pervasive Computing, pp. 201–219, 2014.
[Online]. Available: https://www.igi-global.com/chapter/challenges-
for-personal-data-stream-management-in-smart-buildings/88816
[297]S. Talari, M. Shafie-khah, P. Siano, V. Loia, A. Tommasetti, and J. P. S. Ala Al-Fuqaha (S’00-M’04-SM’09) received Ph.D.
Catalo, “A Review of Smart Cities Based on the Internet of Things degree in Computer Engineering and Networking
Concept,” Energies, vol. 10, no. 4, pp. 1–23, 2017. [Online]. Available: from the University of Missouri-Kansas City, Kansas
https://ideas.repec.org/a/gam/jeners/v10y2017i4p421-d93860.html City, MO, USA, in 2004. His research interests
[298]J. Ye, S. Dobson, and S. McKeever, “Situation identification include the use of machine learning in general and
techniques in pervasive computing: A review,” Pervasive and Mobile deep learning in particular in support of the data-
Computing, vol. 8, no. 1, pp. 36–66, Feb. 2012. [Online]. Available: driven and self-driven management of large-scale
http://www.sciencedirect.com/science/article/pii/S1574119211000253 deployments of IoT and smart city infrastructure and
[299]B. Babcock, S. Babu, M. Datar, R. Motwani, and J. Widom, services, Wireless Vehicular Networks (VANETs),
“Models and Issues in Data Stream Systems,” in Proceedings cooperation and spectrum access etiquette in cogni-
of the Twenty-first ACM SIGMOD-SIGACT-SIGART Symposium on tive radio networks, and management and planning
Principles of Database Systems, ser. PODS ’02. New York, of software defined networks (SDN). He is a senior member of the IEEE
NY, USA: ACM, 2002, pp. 1–16. [Online]. Available: http: and an ABET Program Evaluator (PEV). He serves on editorial boards
//doi.acm.org/10.1145/543613.543615 and technical program committees of multiple international journals and
[300]G. Krempl, I. liobaite, D. Brzeziski, E. Hllermeier, M. Last, conferences.
V. Lemaire, T. Noack, A. Shaker, S. Sievi, M. Spiliopoulou, and
J. Stefanowski, “Open Challenges for Data Stream Mining Research,”
SIGKDD Explor. Newsl., vol. 16, no. 1, pp. 1–10, Sep. 2014. [Online].
Available: http://doi.acm.org/10.1145/2674026.2674028
[301]E. Al Nuaimi, H. Al Neyadi, N. Mohamed, and J. Al-Jaroodi,
“Applications of big data to smart cities,” Journal of Internet Services
and Applications, vol. 6, no. 1, p. 25, Dec. 2015. [Online]. Available:
https://doi.org/10.1186/s13174-015-0041-5
[302]J. Al-Jaroodi and N. Mohamed, “Characteristics and Requirements of Ajay Gupta (S’88 – M’89 – SM’05) received his
Big Data Analytics Applications,” in 2016 IEEE 2nd International Ph.D. in Computer Science from Purdue University
Conference on Collaboration and Internet Computing (CIC), Nov. in 1989, his M.S. in Mathematics and Statistics from
2016, pp. 426–432. the University of Cincinnati in 1984, and his B.E.
[303]P. Thakuriah, N. Tilahun, and M. Zellner, “Big data and urban (Honors) in Electrical and Electronics Engineering
Informatics: innovations and challenges to urban planning and from Birla Institute of Technology and Sciences,
knowledge discovery,” in Seeing Cities Through Big Data: Research Pilani, India in 1982. He is currently a Professor of
Methods and Applications in Urban Informatics, P. Thakuriah, Computer Science at Western Michigan University,
N. Tilahun, and M. Zellner, Eds. New York: Springer, 2016, pp. 11–45. Kalamazoo, MI, USA. From 1998 to 2002, he was
[Online]. Available: http://www.springer.com/gb/book/9783319409009 the Chairman of the Computer Science Department
[304]A. Gandomi and M. Haider, “Beyond the hype: Big data concepts, at Western Michigan University. He has also been
methods, and analytics,” International Journal of Information Man- two term Chair of the IEEE-CS Technical Committee on Parallel Processing
agement, vol. 35, no. 2, pp. 137–144, Apr. 2015. [Online]. Available: from 2011 to 2015 and Vice-Chair of the Technical Activities Committee of
http://www.sciencedirect.com/science/article/pii/S0268401214001066 the IEEE-CS in 2015-2016. His research interests include high performance
[305]K. Bouchard and S. Giroux, “Smart Homes and the Challenges computing, proteogenomics, data analytics, machine learning, sensor systems,
of Data,” in Proceedings of the 8th ACM International Conference cloud computing, mobile computing, web technologies, computer networks,
on PErvasive Technologies Related to Assistive Environments, ser. evolutionary computation, scientific computing, and design and analysis of
PETRA ’15. New York, NY, USA: ACM, 2015, pp. 66:1–66:4. parallel and distributed algorithms. He has published numerous technical
[Online]. Available: http://doi.acm.org/10.1145/2769493.2769519 papers and book chapters in refereed conferences and journals in these areas.
[306]N. Javaid, I. Khan, M. N. Ullah, A. Mahmood, and M. U. Farooq, Dr. Gupta is a senior member of the IEEE and member of the IEEE
“A Survey of Home Energy Management Systems in Future Smart Computer Society, the IEEE Communications Society, the ASEE and the
Grid Communications,” in 2013 Eighth International Conference on ACM. He actively helps organize various ACM and IEEE conferences. He
Broadband and Wireless Computing, Communication and Applications, is also involved in the global efforts to revise undergraduate and graduate
Oct. 2013, pp. 459–464. computer science and computer engineering curriculum to keep pace with the
[307]S. Murphy, This is your future smart home and no, it technological advances.
doesn’t look like ’The Jetsons’, Jan. 2015. [Online]. Available:
https://mashable.com/2015/01/10/future-smart-home/
Driss Benhaddou (S’97 – M’02) received the M.S. Alvis C. Fong (M’97 – SM’04) received the BEng
and two Ph.D. degrees in optoelectronics, engineer- (Hons.) in information systems engineering and MSc
ing, and telecommunications from the University of in electrical engineering from Imperial College Lon-
Montpellier, France, and the University of Missouri- don, England, and PhD in electrical engineering
Kansas City, Kansas City, MO, USA, in 1991, 1995 from University of Auckland, New Zealand.
and 2002, respectively. He is currently an Associate He began his professional career as a software
Professor and the Director of the Wireless and RD Engineer with Motorola. Currently with Western
Optical Networking (WON) Research Laboratory, Michigan University, MI, he has previously held
Department of Engineering Technology, University faculty positions at Massey University, Nanyang
Houston, Houston, TX, USA. He served as the Prin- Technological University, Auckland University of
cipal Investigator (PI) or Co-PI on multiple research Technology, and University of Glasgow, as well as a
projects funded by NSF, NASA, Sprint, ATT, and the University of Houston. visiting position at University of California Irvine. To date, he has published
His research interests include Internet of Things applications to smart systems two books, 13 book sections, and more than 180 papers in leading international
such as smart buildings, smart grid, smart cities, optical networking, sensor journals and conference proceedings, e.g. IEEE T-KDE, IEEE T-AC, IEEE T-
networks, switching system design, routing protocols, performance analysis, II, IEEE T-EC, and contributed to two international patents owned by
and optical instrument development for defect recognition of semiconductors. Motorola. His research interests are in applied AI and data mining for
Dr. Benhaddou has served as a Technical Program Committee Member and knowledge discovery.
a reviewer of many international conferences and journals. He served as Dr. Fong is a Fellow of IET, a Chartered Engineer registered in the UK,
keynote speaker in many international conferences. He organized an NSF and a European Engineer. He has been an Associate Editor of IEEE T-CE
sponsored workshop on wireless application in smart cities in 2016 and co- since 2013.
chaired an IEEE conference on smart cities in 2017. He was the recipient of
the Outstanding Researcher Award at the College of Technology, University
of Houston in 2007.