IUST v26n3p193 en

International Journal of Industrial Engineering & Production Research
September2015, Volume 26, Number 3
pp. 193-211
http://IJIEPR.iust.ac.ir/
pISSN: 2008-4889
A Three- Stage Algorithm for Software

Cost and Time Estimation in Fuzzy
Environment
FatemeZareBaghiabad , Hassan KhademiZare*
Ms.c student yazd university - - yazduni - fatemezarebaghi@yahoo.com
Associated professor yazd university - - yazduni - hkhademiz@yazduni.ac.ir
KEYWORDS
ABSTRACT
software,
cost and time estimation,
fuzzy logic,
network models,
analytical hierarchy process,
The software productions are very expensive projects with

uncertainty. For these reasons, the estimation of time and cost of
software isvery important for both producers and consumers. In this
paper an efficient three- stage algorithm is developed for software
production cost and time estimation. In the first stage, the required
person- month for implementation of software production are
obtained by COCOMO and function point methods. We integrated
these two methods to consider all aspects in software production and
increase estimation accuracy. In the second stage the required
duration for completion of each step of production (planning,
analysis, design and so on) is obtained by paired comparisons matrix.
In third stage, tables of complete time and cost of software are
concluded by GERT network in project control and work break
structure (WBS). In whole of all stages of this paper, triangular fuzzy
numbers are used to express uncertainty existed in succession and
repetition of each production step, time of beginning, ending, the
duration of each task and costs of them. Retrieved results examined by
30 practical projects and conclude accuracy of 93 percent for time
estimation and 92 percent for cost one. Also suggested algorithm is
more accurate than COCOMO 2000 algorithm as 50 percent based
on examined problems.
2015 IUST Publication, IJIEPR, Vol. 26, No. 3, All Rights Reserved.
1. Introduction 1
based on experts opinions who have gained
Generally, there are five methods for project cost
especial
and time estimation. Model-Based method is
projects. Learning-Oriented method estimates
based on mathematic models. Its principals are
project cost and time by simulating previous
mainly
projects.
derived
from
actual
data
sets
of
implemented projects. Expertise-Based method is
skills
over
implementing
Dynamic-Based
method
different
clearly
recognizes characteristics of labor effort, skills,

software project costs and their changes during
Corresponding author:Hassan khademizare

Email:hkhademiz@yazduni.ac.ir
project. Composite method uses the combination
International Journal of Industrial Engineering & Production Research, September 2015, Vol. 26, No. 3
194
FatemeZareBaghiabad , Hassan KhademiZare
A Three- Stage Algorithm for Software Cost and Time. .
of the above methods for the cost and time
Up method begins from components level and
estimation of project[1]. There are different
ends to system. Efforts needed for development of
methods for cost and time estimation of software
any components is computed and then added to
development. The accuracy of the estimation is of
effort costs needed in total system[8]. Case-Based
significant importance for the organization.
Reasoning method belongs to the category of
Primary
on
Machine Learning methods of cost estimation
regression analysis or mathematical derivations
techniques[3]. One of the important steps of this
and current models are based on Simulation,
technique is feature selection which can be highly
Neural Network, Genetic Algorithm and Soft
precise in computational complexities[9]. Neural
Computation[2]. To name a few of the most
Network can model complicate relationships
important methods, Estimation by Analogy,
between
Expert opining, Delphi, Work Break Structure,
independent variables (cost component) and also
Top-Down
Cased-Based
use training records to be generalized for new data
Reasoning, Neural Network, Fuzzy Logic Model,
and then achieve acceptable results[10]. In the
Function
and
same context, by using Probabilistic Neural
COCOMO can be mentioned[3]. Estimation by
Network can simultaneously reach estimations of
Analogy is appropriate when similar implemented
development software parameter (size or effort)
projects are available in the same field[4]. Expert
and probability which real value of parameter is
Opining method is quite useful when there is
less than estimation one[11]. Fuzzy Logic Model
similar
cost
is also categorized under Machine-learning
calculation, an expert system is designed and
techniques. It may begin with Short-Scale
implemented using experts knowledge in similar
programs and the codes collected can be used as
projects.
direct
inputs for a fuzzy model of software development
communications with experts for cost and time
effort estimation[3]. One of the other vastly used
estimation.
when
methods for software cost estimation is Function
organization does not have proper data sets for
Point Analogy based on Source Lines of Code.
cost estimation. WBS method has various
Due to its full quantity control capability at the
applications in general software projects by break
end of estimation, this method is the most
of hierarchical of tasks, systems and subsystems
intuitive one for cost estimation. This method
for scheduling and budgeting[5]. Top-Down
considers the whole software as a set of functions
method begins estimating from system level by
whose quantity determines the approximate
testing
program size[12].
estimation
and
Point
models
Bottom-Up,
Analogy,
historical
of
method
very
product
based
COCOMO
information.
Delphi
Its
are
For
uses
convenient
macro
functions
and
depended
variables
(effort)
and
interrelation between sub functions. The cost of
COCOMO (Constructive Cost Model) approach is
system level activities such as integration,
designed for cost calculation of large software
configuration management and documentation is
systems. COCOMO is a developed version of
estimated and added to accounts[6, 7]. Bottom-
COCOMO.
COCOMO
is
hierarchy
of
A Three- Stage Algorithm for Software Cost and Time. .. . 195
estimation models and is applicable to more
Capabilities,
modern
Capability, Analysis Capability and Product
kinds
of
more
volatility
software
Time
Constraint,
Programmer
development models such as business software,
Complexity[17].
object-oriented software and software which uses
Software cost and time estimation models can also
spiral or additional models[13]. COCOMO uses
be
Function Points or Lines of Code to estimate a
algorithmic
software system size. However, Lines of Code
approaches often estimate cost and time by using
cant
either previous project information or people
be
estimated in the
development
stages[14].
In
early software
COCOMO
and
categorized
into
algorithmic
approaches.
and
non-
Non-algorithmic
comments and experience instead of complex
COCOMO, 12 and 17 system properties are
mathematical
examined,
algorithm
algorithmic approaches have more complex
analyzes characteristics of the four effective
mathematical computations and are based on
factors in software system including product,
mathematical models. These models attempt to
computer, human resource and project. Product
relate effort to one or two project characteristics.
characteristics comprises of ordered software
The main cost component of the model is usually
reliability,
project
software size (such as number of lines of code).
complexity. Computer characteristics include run
General forms of these models are linear or
time standards, main memory standards, boot
nonlinear regression[3].
time,
Personnel
In this paper, we computed required person-
characteristics include analytical capability, work
month for implementation of production projects
plan experience, programming capability, dummy
by developed COCOMO method. In COCOMO
business and programming language experience.
model, network factors including geographical
Project
documented
extends of system, system reliability in critical
programming operations, software tools usage and
conditions, and number of simultaneous users,
constrains on the implementation schedule[16].
accessibility to system support and coordination
There are 22 factors in COCOMO including
level with existing systems have less important; in
Development
Cohesion,
this model less system functions are also used. So
Develop of Reuse capability, Unprecedented,
we added network factors to COCOMO method
Architecture
and
Platform
and named it developed COCOMO. Also we
Experience,
Data
Required
considered number of function points as software
Development Schedule, Language and tools
size in COCOMO model. We raised accuracy of
Experience, Process Maturity, Storage Constraint,
estimation by integrating of these two models and
Use of Software tools, Platform Volatility,
reducing their deficits. We used analytical
Application Experience, Multi-site Development,
hierarchy
Documentation Match to Life Cycle Needs,
determine the influence of each of 33 project
Required
factor
respectively[15].
database
dummy
size,
work
software
capability.
characteristics
contains
Flexibility,
risk
Software
This
Team
Resolution,
Base
size,
Reliability,
Personnel
formula
process
and
(mentioned
and
equations,
expert
in
opinions
but
to
developed
196
COCOMOmodel) on each of 8 production steps
to determine the influence of project sequence and
(Determine required systems, Understand the
reproducibility on project time and cost.
system requirements, analysis, design, system
In
implementation, test and delivery, Maintenance
parameters such as activity definition, sequence
and development, installation) in second stage. To
and time are definitive. In stochastic network only
maintain flexibility of developed algorithm in
parameter of activity time is uncertain estimated
various conditions, the coefficients of model were
by probabilistic distribution. This technique
determined as fuzzy number by fuzzy LIKERT
obtains more real results than certain ones but
spectrum. Coefficients used in this paper are
cant consider cycles and uncertainty in activity
designed for specific place conditions.
definition and sequence. The cyclic stochastic
certain
network
techniques,
network
network with logical nodes, stochastic paths and

1.1-
repetitive loops are appropriate tools for modeling
Production scheduling techniques in
of software production. This technique can
uncertain conditions
Determining
of
tables
of
scheduling
and
consider cycles and uncertainty in activity times.
programming of software production is one of the
Chanas
basic elements of project management of software
Prade[19] were the first to use fuzzy logic to
production. Inappropriate and unreal estimations
schedule project development. Many techniques
of tables of production program make wrong
have been offered for fuzzy scheduling so far
assessment in cost, required resources and
which can be categorized into three categories
scheduling; also leads to customer unreliability,
based on their application: fuzzy time, fuzzy
applying costs in contracts and contradiction in
network and cyclic fuzzy network[20]. In project
projects progress reports. One of the reasons of
schedule technique with fuzzy time, only the
these unreal estimations is using inappropriate
network time parameter is fuzzy. Cyclic Fuzzy
estimation techniques. Software projects have
network are the same cyclic stochastic network
specific properties which distinguish it from other
which fuzzy parameters replace with stochastic
projects and approach software projects to
ones. Itakura and Nishakava[21] in 1984 first
research
encompass
suggested cyclic fuzzy networks. Ozdamar and
uncertainty in definition, sequence and time of
Alanya[22] in 2000 examined software production
activities, required resources, existence of loops,
projects and suggested modeling a problem of
and lack of reproducibility in project activities and
mathematical nonlinear 0-1 programming. Their
so on. Three properties of uncertainty in time of
model includes uncertainty in activity duration
activity, sequence and reproducibility are more
and network architecture. Activities can be done
important than other properties because of their
in parallel. Objective function is to minimize total
direct effect on project required time and
project duration.
ones.
These
properties
resources. We also apply fuzzy GERT technique
and
Kamburowski[18]
and
Henry
In this paper parameters of activity duration and

number of loop iteration were shown as fuzzy
Graphical Evaluation and Review Technique
triangular numbers in GERT networks. Output of
External Output (EO) and External queries (EQ).
this algorithm was shown as a table of production
Each function has three complexity degrees of
programming for all steps of software production
Low, Average and High. Number and weight of
that was depicted with fuzzy numbers. To test and
each function is calculated according to standard
examine validation of this algorithm, thirty real
Tab 1[23]. We adjusted this standard table
software projects were considered and estimated
proportional to expert opinion as fuzzy numbers.
results were compared with real information.
Numbers inside parenthesis shows weights of

number of functions. Each function has three
1.2-
degree of complexity: low, average and high
Functions of Software Systems
shown with fuzzy triangular numbers. For

Function points approach estimates cost by
example, software experts conclude if the
determining of software size. Software systems
numbers of internal logic files are between 10 and
retain five functions: Internal Logic File (ILF),
50, N will be 7 with degree of complexity of 0.9.
External Interface File (EIF), External Input (EI),
Tab 1: number of functions

Low
Average
High
(10, 30, 50)

N= 7
(30, 100, 200)
(100, 300, 500)
ILF
(W=0.9)
N=10(W=1.05)
N=15(W=1.3)
(10, 30, 50)

N=5
(30, 100, 200)
(100, 300, 500)
EIF
(W=0.8)
N=7 (W=0.9)
N=10(W=1.05)
(5, 25, 40) N=3
(25, 75, 125)
(100, 200, 300)
EI
(W=0.7)
N=4 (W=0.75)
N= 6(W=0.85)
(5, 25, 40)

N=4
(25, 75, 125)
(100, 200, 300)
EO
(W=0.75)
N=5 (W=0.8)
N=7(W=0.9)
(10, 30, 50) N=3
(30, 80, 130)
(100, 200, 300)
EQ
(W=0.7)
N=4 (W=0.75)
N=6(W=0.85)
W ij = Weight of number of functions type i with
As a result, to estimate software size, in addition
Group
to the number of functions, their complexities

must also be estimated. Software size is named as
unadjusted function point (UFP) and computed
complexity level j
UFP = Number of unadjusted functions
We use UFP as size of software to estimate
according to the Eq.1[5]:
required person- month for implementation of

5
ij
UFP =
production projects.
N W
i-1
ij
j-1
(1)
ij
i-1
2. Effective factors for software cost and
j-1
time estimation
N ij = Number of functions type i with complexity
In primary computers, software costs constituted a
level j
minor percentage of total computer system costs,
198
thus some errors in software cost estimation had
Developed model has an exponential factor for the
relatively low effect on total system cost, but
most
nowadays software is the most expensive
conditions. Effective factors in determining power
component in computers so that error of
of weights of information systems (WIS) include
estimation can be equal to the gap between benefit
available
and loss.
project
Cost and time estimation is usually fuzzy and an
integrity and project control. Each factor has five
imprecise science. Countless variables such as
states which are effective in determining WIS
personnel, engineering issues, environmental
power: very high, high, medium, low and very
conditions and government policies can affect
low.
probable,
optimistic
and
pessimistic
experience,
production
flexibility,
management
capability,
teamwork
final software cost and the work done for

development. The main component of software
development cost is due to man and his efforts.
Most cost estimation methods focus on this aspect
and estimate costs as person-month phrase. In
order to achieve a reliable cost and work
estimation of a software project, either a set of
systematic steps must be designed or the software
must be broken down into different functional
components so that to have a sound estimation
with an acceptable level of risk.
costs and the final price that the costumer is

charged. This is because it depends on different
factors including organization validity, travels,
required trainings, software engineers salary,
project characteristics, personnel, network and
hardware which last during project life.
product characteristics first, and then minimize

error probability by using numbers and cost
this
method,
project
cost
estimation is often implemented parallel to the

project
planning
conceivable
step
project
so
characteristics
of
information technology systems

There are generally 33 cost and time factors
which determine software production time and
cost. These factors are categorized into five
categories: characteristics of product, computer,
personnel, project and network. The coefficient
mentioned in this table is determined based on
condition of production system and characteristics
that
to
characteristics
coefficients are summarized is in Tab2. The

coefficients of Tab2 are obtained based on field
studies,
experts
sessions
and
practical
experiences available in three major centers of

design, production and support. Each question in
this table has five states: very high, high, medium,
low and very low. Due to the diversity of
The models practitioners must properly recognize
In
and
of five above factors. Required information and
There is not a simple relationship between project
coefficients.
2.1-Factors
achieve
and
incorporate sound budget numbers into the model.
questions, in some cases very much and in some

cases very less is appropriate.
Selecting any of the state influence software time
and cost estimation. For an even more accurate
estimation, at least 5 experts are required to fill
this table. The mean score of the experts scores
will be the basis of time and cost computation. To
perception details of Tab2, see appendix A.
A Three- Stage Algorithm for Software Cost and Time. . .
199
Tab 2: Summery of information and coefficients
S8
C1
C2
C3
C4
C5
H1
H2
H3
H4
H5
P1
P2
P3
P4
P5
P6
P7
P8
P9
P10
N1
N2
N3
N4
N5
Product
Characteristics (S)
Computer
Characteris
tics (C)
S7
Personnel
Characteristics (H)
S6
Project
Characteristics (P)
S5
Title
Network
Characteris
tics (N)
Row
S1
S2
S3
S4
Cost Factors
Database Size
Complexity of program structure
Flexibility & predictability capability
Importance of user information security
Importance of compatibility with
relevant applications (system
Integration)
Observation of software standards
Usage of modern software
tools(required)
Number of inputs and outputs
Required memory to run
Compatibility with existing hardware
Computer response time(run time)
Usage of modern hardware tools
Diversity of existing hardware
Analyst group capability ( infer &
provide requirements)
Designer group experiences about
considered system
Programmer group capability
Group knowledge about present
program(hardware& software)
Group capability in system integration
Usage by professional programmers
Project scheduling changes
Project delivery time
System validity and verification
Software response level
Ease of Install & launch
Software documentations
Support Period ( maintenance,
modification, user training)
Number of main software functions
system development capability
Geographical Extends of system
Number of simultaneous users
Access to system support
Reliability in Critical situations
Coordination with present systems
Very high
1.86
1.3
1.5
1.4
High
1.28
1.15
1.3
1.15
Average
1.08
1
1.1
1
Low
0.94
0.85
1
0.88
Very low
0.5
0.7
0.8
0.75
1.3
1.2
1.1
0.8
1.1
1.05
0.7
0.3
1.5
1.3
1.3
0.8
0.5
1.4
1.21
1.3
1.15
1.3
1.1
1.3
1.06
1.15
1.07
1.11
1
1.3
1
1
1
1
0.9
0.9
0.5
0.87
0.87
0.5
0.7
0.7
0.25
0.3
0.35
0.25
0.3
0.71
0.86
1.19
1.46
0.82
0.91
1.13
1.29
0.7
0.86
1.19
1.46
0.5
0.9
1.1
1.2
0.5
0.7
1.1
0.82
1.3
0.7
1.3
1.3
0.95
0.86
1.04
0.91
1.11
1
1.1
1.1
1
1
1
1
1
1.1
1
1
1.07
1.17
1.08
1.12
1.11
1.2
0.8
1.1
1.14
1.42
1.23
1.3
1.3
1.5
0.5
0.5
1.5
1.3
1.2
0.8
0.5
1.4
1.3
1.5
1.4
1.4
1.3
1.4
1.3
1.11
1.3
1.15
1.2
1.15
1.3
1.2
1.05
1.2
1
1.1
1
1.1
1
0.5
0.7
0.58
0.7
0.87
0.8
0.4
0.25
0.5
0.3
0.5
0.3
0.25
In this step, first Tab1 is completed by n experts.
3.
First Stage of Hybrid Algorithm:
Time and Cost Estimation Algorithm
Weight average of function points (UFP) is

calculated as Eq.1. If Eq.1 is completed by more
benefits,
one expert UFP will compute by average of all
shortcomings and applications of each method of
UFPs. Our purpose of n experts is that people who
software cost estimation, our algorithm steps
are aware about software functions and work
which are a combination of Functional and
practical software production projects.
COCOMOtechniques are as follows:
Step 2: Determine each coefficient of Tab2. (n
Step 1: Calculate software size by function points
experts must complete this table).
approach (UFP).
Step 3: Multiply the resulted weights in all rows
After
studying
and
examining
ofEach coefficient of Tab2 selected by an expert
200
is called effort multiplier (EM). The obtained
month
for
project
implementation.
These
number is weights of information systems (WIS)
equations are the same standard relations in
and obtained as Eq.2 [24]. If Tab2 is completed
COCOMO [25].
by more one expert, average of WIS for all

experts must be computed.
33
WIS = EM
(2)
i =1
Person monthmin = 2.8(WIS )(UFP)1.05
(3)
Person monthnom = 3(WIS )(UFP)1.12
(4)
Person monthmax = 3.2(WIS )(UFP)1.2
(5)
Step 4: Calculate the minimum, maximum and

most likely project duration by Eq.3, Eq.4 and
Eq.5. Resulted numbers express required person-
Step 5: Distribute each of the calculated times

between project personnel according to Tab3.
Tab 3: Staff cost and time percentage

Personnel
Characteristics
Time percentage
Firm
management
4.9, 5, 5.1
Project
management
14.8, 15, 15.2
MA
Undergraduate
Technician
24.7, 25, 25.3
19.8, 20, 20.2
34.5, 35, 35.5
General objective of problem is in head of

Second stage of hybrid algorithm:
this process and criteria, sub criteria and
Paired comparisons matrix of analytical
alternatives are in next levels. Qualitative and
hierarchy process (AHP) for determining
quantitative objectives are measured as paired
relationship between factors and steps
comparisons for comparing alternatives to
The world around us is full of multi- criteria
criteria and criteria to objectives[26]. We
problems which we must make decision. One
used the fuzzy spectrum with five options in
of the most complete approaches for making
these paired comparisons as Tab 4 and Fig. 1.
decision is analytical hierarchy process which
In this paper, we assumed factors (product,
first suggested by Saaty in 1980. This process
project, computer, personnel and network) are
is a graphic view from real complex problem.
independent from each other.
4.
Tab 4: The relationship for paired comparisons

Alternative
Excellent
Good
Fair
Poor
Very Poor
Alternative weight
(0.75, 1, 1)
(0.5, 0.75, 1)
(0.25, 0.5, 0.75)
(0, 0.25, 0.5)
(0, 0, 0.25)
Symbol
E
G
F
P
V
Linguistic phrase
Very high
High
Medium
Low
Very low
Fig.1: Review of fuzzy numbers

Early weights of alternatives to criteria (steps) and
experts. To collect information we design a
criteria to general objective were calculated by
questionnaire with three sections. In first section
dividing retrieving weights on each factor of
we determine importance of each alternative to
questionnaire (
wi
) and comparing every element
wj
factors.
In
second
section
we
determine
importance of factors to steps and in third section
of each level to other levels. Final weight of each
determine the influence of each step on software
criterion and alternatives to each other was
cost and time estimation. Related experts were
calculated
hierarchy
from three major centers of design, production
process and Expert Choice (EC) software. Chart
and support. These experts were capable and
of analytical hierarchy process was shown as
possessor of at least three years work experience;
Figure 2 and calculated weights depicted as Tab
thus in this paper statistical sample was equal to
5.
statistical society.
by
applying
analytical
To increase accuracy and validity of this

algorithm, we used opinions of 150 software
Fig.2: Analytical hierarchy process

202
Tab 5: Time percentages of steps

The influence of each factor on each production step
Factors:
Product
Computer
Personnel
Project
Network
Planning
Understan
Determin
d the
e required
system
systems
requireme
nts
Implement
Analysis
Design
System
implement
Test and
Deliver
Maintenan
ce and
developm
ent
Installati
on
Time
percentage
2.9, 3, 3.1
19.9, 20,
20.1
16.8, 17,
17.2
24.8, 25,
25.2
23.8, 24,
24.2
10.8, 11,
11.2
4.9, 5,
5.1
4.9, 5, 5.1
Support
Formal Methods model and Fourth Generation

Total duration of project was calculated by
techniques[27]. In this paper in order to analyze
following steps 1 to 3 in explained algorithm.
and evaluate the developed algorithm, Waterfall
Then the duration of each production step was
method is used.
obtained by multiplying total duration to time

percentage of each step. The durations of each
5. Third stage of hybrid algorithm:
step were inputs of third stage of the algorithm. In
Development of solving method for GERT
third stage total time and cost of project was
networks in uncertain condition
obtained by calculations of fuzzy GERT networks
After determining the duration of each step of
and development algorithm.
software production, we must define nodes, paths

and production loops for reaching tables of
4.1-
production programming. To model software
System development steps
Cycle of Software system development has
projects explained in this paper, we use cyclic
different methodologies in macro levels. The most
fuzzy graphical evaluation and review technique
important
is
(GERT) networks which parameters are shown as
cycle
fuzzy sets[28]. First projects information were
(ISDLC), since it has numerous applications and
estimated based on definitions and assumptions,
encompasses all steps of planning, analysis,
then network of software steps were depicted and
design, implementation and software support.
in next step cyclic fuzzy network was solved.
Different models are designed for implementing
These steps were explained in[28]. Output of this
information system development life cycle steps.
algorithm
The most important methodologies used in
programming,
software production include: Linear Sequential
completion time of project as fuzzy numbers.
and
information
complete
system
methodology
development
life
includes
tables
of
implementation
production
steps
and
model, Waterfall model, Prototyping model,

Rapid
Application
Incremental
model,
Development
Spiral
model,
model,
5.1-
Parallel
programming
Development, Object Oriental Component model,
Analysis of results of production
Table of production programming including
implementation time of each step of software
implementation by Eqs.1, 2 and 3:
production obtains by evaluating each node and
Person-Month= (person-month
activity. Since calculations are based on fuzzy
Nom,
numbers, in cyclic fuzzy networks the completion
Step4. Compute time percent of each production
time of project is a fuzzy number instead of a
step by last row of Tab 5: For this purpose, values
definitive one. This uncertain numbers adapt to
of person- month were multiplied by values of last
reality and decreases stresses in project control
row of Tab 5 and results have been in second
sessions. Also the cost is estimated more real.
column of Tab 6. In Tab 6, number and
Computational model developed in this paper is
probability of loop were determined by experts.
designed in Excel environment and simply
Step5. Do network computations and complete
extended for other software. This model is
Tab.7 and Tab.8: These computations were done
capable of upgrading in new conditions. Fuzzy
based on computations of paper [28]. The
GERT chart used in this paper depicted steps of
completion time of each production step and
design, production and software support with
completion time of project was determined by
loops is shown in Fig. 3.
Tab.7 and Tab.8. The completion time of project
Min,
person-month
person-month Max)
was computed by Eq.6:

5.2-
Validation and verification of

The completion time =
developed algorithm
To examine validity and verification, 30 real
projects were considered implemented in three
(F8min + (F8nom * 4) +
(6)
F8max) / 6
major centers of design, production and support.

Time and cost of these projects were estimated by
Step6. Compute project cost by Tab. 9: In this
of
table, participation percent and salary of each
COCOMO2000 and then compared with real
work group affected on software production was
time and cost. Steps of cost and time estimation
determined and by help of this, software cost was
for software were explained as below:
computed.
Step1. Determine value of UFP by Tab1 and
Step7. Compare estimated time and cost of
expertise comments
software production with real ones: values of last
Step2. Complete Tab2 by expertise comments and
row of Tab 10 expressed improvement on
compute WIS
production time and cost of software in suggested
Step3. Compute minimum, maximum and most
algorithm relative to real values and COCOMO
likely values of required person- month for project
2000 method.
developed
algorithm
and
method
204
Fig.3: model of software production

Tab 6: Network loops and activities
Activity
code
Rank
(loop)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
1
2
3
4
5
6
7
8
21
32
43
54
65
76
87
Activity description (loop)

Understanding system needs and requirements
Determining required systems
System analysis
System design
System implementation
System setup and installation
System test and delivery
System maintenance and support
Loop 2 1
Loop 3 2
Loop 4 3
Loop 5 4
Loop 6 5
Loop 7 6
Loop 8 7
Activity duration
(month)
Loop frequency
(number)
(a1, b1, c1)
(a2, b2, c2)
(a3, b3, c3)
(a4, b4, c4)
(a5, b5, c5)
(a6, b6, c6)
(a7, b7, c7)
(a8, b8, c8)
(2, 3, 4)
(1, 2, 3)
(1, 2, 3)
(2, 3, 4)
(3, 4, 5)
(1, 2, 3)
(2, 3, 4)
Activity
occurrence
probability
(loop)
1
1
1
1
1
1
1
1
0.14
0.11
0.12
0.11
0.15
0.14
0.12
Tab.7: Activity duration including iteration loops

Activity
Iteration
loops
Loop
occurrence
probability
21
0.14
(a1, b1, c1)
(2, 3, 4)
(0.3, 0.4, 0.6)
(1.3, 1.4, 1.6)
21
0.14
(a2, b2, c2)
B1=
1.3a1, 1.4 b1, )
(1.6c1
(2, 3, 4)
(0.4, 0.6,
(1.4, 1.6, 1.9)
B2=
Activity
duration
Frequency
loop
Activity
frequency
Final activity
frequency
Final activity
completion time

32
0.11
32
43
0.11
0.12
43
54

(1, 2, 3)
0.s9)
(a3, b3, c3)
(1, 2, 3)
(1, 2, 3)
(0.2, 0.5, 0.7)
(1.2, 1.6, 1.7)
0.12
0.11
(a4, b4, c4)
(1, 2, 3)
(2, 3, 4)
(0.3, 0.6, 0.8)
(1.3, 1.6, 1.8)
54
65
0.11
0.15
(a5, b5, c5)
(2, 3, 4)
(3, 4, 5)
(0.7, 0.9, 1.2)
(1.7, 1.9, 2.2)
65
76
0.15
0.14
(a6, b6, c6)
(3, 4, 5)
(1, 2, 3)
(0.6, 0.9, 1.2)
(1.6, 1.9, 2.2)
76
87
0.14
0.12
(a7, b7, c7)
(1, 2, 3)
(3, 4, 5)
(0.5, 0.8, 1)
(1.5, 1.8, 2)
87
0.12
(a8, b8, c8)
(3, 4, 5)
(0.4, 0.5, 0.6)
(1.4, 1.5, 1.6)
1.4a2, 1.6 b2, )

(1.9c2
B3=
1.2a3, 1.6 b3, )
(1.7c3
B4=
1.3a4, 1.6 b4, )
(1.8c4
B5=
1.7a5, 1.9 b5, )
(2.2c5
B6=
1.6a6, 1.9 b6, )
(2.2c6
B7=
1.5a7, 1.8 b7, 2 )
(c7
B8=
1.4a8, 1.5 b8, )
(1.6c8
Tab.8:Software production schedule

Node
1
2
3
4
5
6
7
8
Input activity
Understanding system needs
and requirements
Determining required systems
System analysis
System design
System implementation
System setup and installation
System test and delivery
System maintenance and
support
Final activity completion
Activities begin time
Activities finish time
B1
R1 = (0,0,0)
F1 = R1 +B1
B2
B3
B4
B5
B6
B7
R2 = B1 + R1
R3 = B2 + R2
R4 = B3 + R3
R5 = B4 + R4
R6 = B5 + R5
R7 = B6 + R6
F2 = R2 + B2
F3 = R3 + B3
F4 = R4 + B4
F5 = R5 + B5
F6= R6 + B6
F7 = R7 + B7
B8
R8 = B7 + R7
F8 = R8 + B8
Tab.9: cost of software production

Required time
for project
completion
B1 + + B8
Sum up monthly
time
Monthly salary
(monetary unit)
The cost of
working group
Sum up total
project costs
5.3-
Percent participation in project time (Error! Not a valid result for table.)
Company
manager
(4.9, 5, 5.1)
(B1+ + B8)*
(4.9, 5, 5.1)
Project
manager
(14.8, 15, 15.2)
(B1+ + B8)*
(14.8, 15, 15.2)
MA
(24.7, 25, 25.3)
BA
(19.8, 20, 20.2)
Associate degree
(34.5, 35, 35.5)
(B1+ + B8)*
(24.7, 25, 25.3)
(B1+ + B8)*
(19.8, 20, 20.2)
(B1+ + B8)*
(34.5, 35, 35.5)
(B1+ + B8)*
(B1+ + B8)*
(B1+ + B8)*
(B1+ + B8)*
(14.8, 15, 15.2)*
(24.7, 25, 25.3)*
(19.8, 20, 20.2)*
(34.5, 35, 35.5)*
B
C
D
E
(B1+ + B8)*[(4.9, 5, 5.1)* A+ (14.8, 15, 15.2)* B+ (24.7, 25, 25.3)* C+ (19.8, 20, 20.2)*
D+(34.5, 35, 35.5)* E]
(B1+ + B8)*
(4.9, 5, 5.1)* A
Compare estimated results with real
information
Time and cost for 30 projects were estimated by
two methods of COCOMO 2000 and algorithm

developed in this paper. Estimated results were
compared with real time and cost as Tab 10, Fig.
4 and Fig. 5. For comparing time and cost
206
software, Eq. 7 was used.
MRE =
| Estimated Actual |
Actual
(7)
Tab 10: compare estimated results with real information

Project
P1
P2
P3
P4
P5
P6
P7
P8
P9
P10
P11
P12
P13
P14
P15
P16
P17
P18
P19
P20
P21
P22
P23
P24
P25
P26
P27
P28
P29
P30
Sum
Average
Real cost and

time
Cost
Time
COCOMO2000 method
Cost
(monetary
unit)
(Personhour)
(monetary
unit)
125
175
280
135
920
825
740
150
200
120
560
565
420
310
650
340
595
780
675
480
350
950
160
890
850
195
215
200
225
180
13260
442
1300
1700
2950
1420
9800
8500
7500
1450
1900
1300
5700
5450
4150
3200
6400
3300
5900
7900
6800
5100
3400
9650
1700
8700
8650
2600
2850
2670
2980
2400
137320
4577.3
144
191
347
151
970
743
703
165
156
111
672
706
370
242
552
306
660
880
580
422
410
810
176
935
757
175
190
220
205
220
13167.5
438.92
Percent
deviation
0.15
0.09
0.24
0.12
0.06
(0.9)
(0.05)
(0.1)
(0.23)
(0.08)
0.2
0.25
(0.12)
(0.22)
(0.15)
(0.1)
0.11
0.15
(0.14)
(0.12)
0.17
(0.15)
0.1
0.05
(0.11)
0.1
0.12
(0.1)
0.08
(0.18)
4.74
0.16
Time
(Personhour)
1450
1870
3540
1620
10585
7820
6975
1625
1520
1015
6670
6705
3735
2435
5440
2937
6726
8850
5985
4590
3875
8396
1887
9405
7612
2540
2710
3020
2900
2970
137408
4580.3
Percent
deviation
0.12
0.1
0.2
0.14
0.08
(0.08)
(0.07)
0.12
(0.2)
(0.22)
0.17
0.23
(0.1)
(0.24)
(0.15)
(0.11)
0.14
0.12
(0.12)
(0.1)
0.14
(0.13)
0.11
0.09
(0.12)
0.02
0.05
(0.13)
0.03
(0.24)
3.86
0.13
Hybrid / developed method

Cost
(monetary
unit)
139
190
311
146
966
756
718
178
180
114
627
635
400
273
579
326
625
855
601
446
392
855
173
917
807
188
204
207
213
199
13219
440.63
Percent
deviation
0.11
0.05
0.11
0.08
0.05
(0.06)
(0.03)
0.08
(0.1)
(0.05)
0.12
0.12
(0.05)
(0.12)
(0.11)
(0.04)
0.06
0.1
(0.11)
(0.07)
0.12
(0.1)
0.08
0.03
(0.05)
0.04
0.05
(0.03)
0.05
(0.1)
2.27
0.08
Time
(Personhour)
1405
1768
3245
1534
10290
8075
7275
1551
1691
1144
6270
4796
3942
2848
5888
3200
6372
8295
6324
4845
3740
8685
1785
8352
7698
2513
2753
2775
2853
2669
134581
4486
Percent
deviation
0.08
0.04
0.1
0.08
0.05
(0.05)
(0.03)
0.07
(0.11)
(0.12)
0.1
0.12
(0.06)
(0.11)
(0.08)
(0.03)
0.08
0.05
0.07
0.05
0.1
0.1
0.05
0.04
0.11
0.03
0.03
(0.04)
0.04
(0.11)
2.13
0.07
1200
1000
Real Cost
800
600
Cost by
COCOMO2000
400
Cost by hybrid
algorithm
200
0
0 1 2 3 4 5 6 7 8 9101112131415161718192021222324252627282930
Fig.4: Comparing real cost with cost estimated by COCOMO2000 and hybrid algorithm
12000
10000
8000
Real Time
6000
Time by
COCOMO2000
4000
Time by hybrid
algorithm
2000
0
0 1 2 3 4 5 6 7 8 9101112131415161718192021222324252627282930
Fig.5: Comparing real time (personhour) with time estimated by COCOMO2000 and hybrid
algorithm
6. Conclusions and Future Researches
to suggested algorithm. The accuracy of 93 and 92
In this paper a three- stage algorithm was
for estimation of software time and cost with high
developed to estimate cost and time of software
uncertainty is very worthy.
production projects in uncertain conditions. First
As future researches, we suggested:
stage includes obtaining required person- month
1.
for implementation of production projects by
systems, we used methods of COCOMO and
combining of two models of function points and
Function points. Other methods can be suggested
COCOMO. In second stage time relationship
to distinguish important specifications which
between specifications of five factors (product,
effect on time and cost estimation.
computer, personnel, project and network) and
2.
eight
were
network was used only for the competition time
determined by paired comparisons matrix of AHP
and then we estimated cost based on the time.
model.
Other approaches can be suggested to estimate
In this stage, the duration of each step was
cost directly by network and fuzzy GERT
determined for iterance to third stage. In third
calculations.
stage, the final time and cost was computed by
3.
work break structure and fuzzy GERT networks.
estimate software sales and its effect on cost and
Results of comparing real information of 30
profitability of software.
steps
implemented
of
software
projects
with
production
estimations
To determine specifications of software
In this paper calculation of fuzzy GERT
Its useful to design a method which can
of
Resources
suggested algorithm express accuracy of 93

percent for time estimation and 92 percent for cost
[1]
Dolado, J.J., "On the Problem of the
one. The maximum relative error for when only
Software Cost Function", Inform Software
COCOMO 2000 method used, is twice relative
Tech, 43 (1), 2001, pp.61- 72.
208
[2]
Choudhary, K., "GA Based Optimization of
[9]
Li, Y.F., Xie, M., Goh, T.N., "A study of
Software Development Effort Estimation",
mutual information based feature selection
International Journal of Computer Science
for case based reasoning in software cost
and Technology, 1 (1), 2010, pp.38-40.
estimation",
Expert
Systems
with
Applications, 36 (3), 2009, pp.59215931.

[3]
Lopez-Martin, C., "A fuzzy logic model for

predicting the development effort of short
[10] Bhatnagar,
R.,
"Software
Development
scale programs based upon twoindependent
Effort Estimation Neural Network Vs.
variables", Applied Soft Computing, 11(1),
Regression
2011, pp.724732.
International
Modeling
Journal
of
Approach",
Engineering
Science and Technology, 2 (7), 2010,

[4]
Idri A., K.T.M. Fuzzy Analogy: A New
pp.2950-2956.
Approach for Software Cost Estimation. in

International
workshop
on
software
[11] Pendharkar, P.C., "Probabilistic estimation

of software size and effort", Expert Systems
measurment. 2001. Aachen, Germany.
with Applications, 37 (6), 2010, pp.44354440.

[5]
Ghahramani,
B.,
"Software
Reliability
Analysis: a Systems Development Model",

Computers and Industrial Engineering, 45
(2), 2003, pp.295-305.
[12] Musilek P., P.W., Succi G., Reformat M.,

"Software Cost Estimation with Fuzzy
Models", Applied Computing Review, 8
(2), 2000, pp.24-29.
[6]
XU, Z., Khoshgoftoar, T.M., "Identification

of
Fuzzy
Models
of
Software
Cost
[13] Dickson, G., Software Cost Estimation
Estimation", Fuzzy Sets and Systems, 145
Canada: Bericht, Faculty of Computer
(1), 2004, pp.141-163.
ScienceFaculty of Engineering, University

of New Brunswick. 2007.
[7]
Huang X., H.D., Ren J., Capretz L.F.,

"Improving the COCOMO model using a
neuro-fuzzy
approach",
Applied
Soft
Computing, 7, 2007, pp.29-40.
[14] Ali A., Q.S., Shah Muhammad S., Abbas J.,

TariqPervaiz M., Awan S., "Software cost
estimation
through
entity
relationship
model", J. Am. Sci., 6 (11), 2010, pp.47-51.

[8]
Izyumov
B.,
K.E.,
Wagenkecht
M.,
Software Tools for Regression Analysis of
[15] Boehm B., A.C., Brown A. W., Chulani S.,
Fuzzy Data, in Proceeding of 9th Zittau
Clark B.K., Horowitz E., Steece B. ,
Fuzzy Colloquium. 2001. p. 221-229.
Software Cost Estimation with COCOMO

IIPrentice- Hall: 2000.
[16] Kazemifard, M., Zaeri, A.,

Aghaee,
N.,
Ghasem-
Nematbakhsh,
M.A.,
[23] Al-Hajri,
M.A.,
Sulaiman,
Abdul
M.N.,
Ghani,
Selamat,
A.A.,
M.H.,
Mardukhi, F., "Fuzzy Emotional COCOMO
"Modification of standard Function Point
II Software Cost Estimation (FECSCE)
complexity weights system", The Journal of
using Multi-Agent Systems", Applied Soft
Systems
Computing, 11 (22602270), 2011.
pp.195206.
and
Software,
74(2),
2005,
[17] Sicila M.A., G.E., Galvo T., "An Inqairy-
[24] Boehm, B., Valerdi, R., lane, J.A., Winsor
based Method for Choquet Integral-based
Brown, A., "COCOMO Suite Methodology
Aggregation
and
of
Interface
Usability
Evolution,
Software
Engineering
Parameters", Kybernetika, 39 (5), 2003,
Technology", Journal of Defense Software
pp.401- 414.
Engineering, 2005, pp.20-25.
[18] Chans., K.J., "The use of Fuzzy Variables in
[25] Jung, P., ed. Software Cost Estimation and
pert", Fuzzy Sets and Systems, 5 (1), 1981,
COCOMO II. 1997, Systems Engineering
pp.11-19.
Research Institute (SERI): Taejon, Korea.
[19] Prade.H, "Using Fuzzy set theory in a

Scheduling Problem: A Case Study", Fuzzy
[26] Ataii, M., Fuzzy multi- critria making

decision Iran: Damghan university. 2011.
sets and Systems, 2 (2), 1979, pp.153-165.

[27] Kargari M., K.H., Information Systems
[20] Wang, J., "A fuzzy robust scheduling
approach
for
product
development
TechnologyIran:
Elmosanat
university.
2005.
projects", European Journal of Operational

Research, 152 (1), 2004, pp.180194.
[28] Khademi Zare, H., Fatemi Ghomi, S.M.T.,

Karimi,
B.,
"Developing
heuristic
[21] Itakura, H., Nishikawa, Y., "Fuzzy Network
algorithm for order production planning
technique for technological Forcasting",
using network models under uncertainty
Fuzzy Sets and Systems, 14 (2), 1984,
conditions",
pp.99-131.
Computation, 182 (2), 2006, pp.12081218.
[22] Ozdamar,
L.,
Alanya,
"Uncertainty
Applied
Mathematics
and
Appendix A
modeling in software development projects
Computation
(with case study)", Annals of Operations
component of Tab.11 is ranking by n experts
Research, 102 (1-4), 2000, pp.157178.
based on the importance of them for estimating
of
coefficients
of
Tab2:Each
time and cost. The concept of each ranking is
210
shown in Tab.12. For each component of Tab2,
resulted numbers are the same coefficients of
very high, high, average, low and very low
Tab2.
coefficients are computed based on Eq. 8- 12. The
S3
S4
S5
S6
Title
Product
Characteristics (S)
Row
S1
S2
C2
C3
C4
C5
H1
H2
H3
H4
H5
Personnel
Characteristics (H)
S8
C1
Computer
Characteristic
s (C)
S7
P8
P9
P10
N1
N2
N3
N4
N5
Network
Characteris
tics (N)
P2
P3
P4
P5
P6
P7
Project
Characteristics (P)
P1
Tab. 11: Computation of coefficients of Tab2

Cost Factors
Importance of each component
Database Size
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Complexity of program structure
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Flexibility & predictability
2 1.75 1.5 1.25 1 0.75 0.5 0.25
capability
Importance of user information
2 1.75 1.5 1.25 1 0.75 0.5 0.25
security
Importance of compatibility with
relevant applications (system
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Integration)
Observation of software standards
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Usage of modern software
2 1.75 1.5 1.25 1 0.75 0.5 0.25
tools(required)
Number of inputs and outputs
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Required memory to run
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Compatibility with existing
2 1.75 1.5 1.25 1 0.75 0.5 0.25
hardware
Computer response time(run time)
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Usage of modern hardware tools
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Diversity of existing hardware
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Analyst group capability ( infer
2 1.75 1.5 1.25 1 0.75 0.5 0.25
&provide requirements)
Designer group experiences about
2 1.75 1.5 1.25 1 0.75 0.5 0.25
considered system
Programmer group capability
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Group knowledge about present
2 1.75 1.5 1.25 1 0.75 0.5 0.25
program(hardware& software)
Group capability in system
2 1.75 1.5 1.25 1 0.75 0.5 0.25
integration
Usage by professional
2 1.75 1.5 1.25 1 0.75 0.5 0.25
programmers
Project scheduling changes
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Project delivery time
2 1.75 1.5 1.25 1 0.75 0.5 0.25
System validity and verification
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Software response level
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Ease of Install & launch
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Software documentations
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Support Period ( maintenance,
2 1.75 1.5 1.25 1 0.75 0.5 0.25
modification, user training)
Number of main software functions 2 1.75 1.5 1.25 1 0.75 0.5 0.25
system development capability
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Geographical Extends of system
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Number of simultaneous users
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Access to system support
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Reliability in Critical situations
2 1.75 1.5 1.25 1 0.75 0.5 0.25
Coordination with present systems
2 1.75 1.5 1.25 1 0.75 0.5 0.25
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Tab. 12: The concept of ranking of Tab.11

Importance
Ranking
Absolutely low
0
Ultra low
0.25
Very low
0.5
Low
0.75
Normal
1
High
1.25
Very high
1.5
Ultra high
1.75
Absolutely high
2
Very high = Max (comments experts)
High= AVE (comments higher than the average)
Average = AVE (comments experts)
Low = AVE (comments lower than the average)
Very low = Min (comments experts)
(8)
(9)
(10)
(11)
(12)

IUST v26n3p193 en

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

IUST v26n3p193 en

Hochgeladen von

Copyright:

Verfügbare Formate

International Journal of Industrial Engineering & Production Research

September2015, Volume 26, Number 3

A Three- Stage Algorithm for Software

The software productions are very expensive projects with

based on experts opinions who have gained

Generally, there are five methods for project cost

and time estimation. Model-Based method is

projects. Learning-Oriented method estimates

based on mathematic models. Its principals are

project cost and time by simulating previous

implemented projects. Expertise-Based method is

recognizes characteristics of labor effort, skills,

Corresponding author:Hassan khademizare

project. Composite method uses the combination

FatemeZareBaghiabad , Hassan KhademiZare

A Three- Stage Algorithm for Software Cost and Time. .

of the above methods for the cost and time

Up method begins from components level and

estimation of project[1]. There are different

ends to system. Efforts needed for development of

methods for cost and time estimation of software

any components is computed and then added to

development. The accuracy of the estimation is of

effort costs needed in total system[8]. Case-Based

significant importance for the organization.

Reasoning method belongs to the category of

Machine Learning methods of cost estimation

regression analysis or mathematical derivations

techniques[3]. One of the important steps of this

and current models are based on Simulation,

technique is feature selection which can be highly

Neural Network, Genetic Algorithm and Soft

precise in computational complexities[9]. Neural

Computation[2]. To name a few of the most

Network can model complicate relationships

important methods, Estimation by Analogy,

Expert opining, Delphi, Work Break Structure,

independent variables (cost component) and also

use training records to be generalized for new data

Reasoning, Neural Network, Fuzzy Logic Model,

and then achieve acceptable results[10]. In the

same context, by using Probabilistic Neural

COCOMO can be mentioned[3]. Estimation by

Network can simultaneously reach estimations of

Analogy is appropriate when similar implemented

development software parameter (size or effort)

projects are available in the same field[4]. Expert

and probability which real value of parameter is

Opining method is quite useful when there is

less than estimation one[11]. Fuzzy Logic Model

is also categorized under Machine-learning

calculation, an expert system is designed and

techniques. It may begin with Short-Scale

implemented using experts knowledge in similar

programs and the codes collected can be used as

inputs for a fuzzy model of software development

communications with experts for cost and time

effort estimation[3]. One of the other vastly used

methods for software cost estimation is Function

organization does not have proper data sets for

Point Analogy based on Source Lines of Code.

cost estimation. WBS method has various

Due to its full quantity control capability at the