SOFTWARE COST ESTIMATION USING FUZZY NUMBER AND PARTICLE SWARM OPTIMIZATION

International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 1, February 2016
DOI: 10.5121/ijci.2016.5112 113
SOFTWARE COST ESTIMATION USING FUZZY
NUMBER AND PARTICLE SWARM OPTIMIZATION
Divya Kashyap1
and D. L. Gupta2
1
Computer Science and Engineering Department, IADC, Bangalore
2
Computer Science and Engineering Department, KNIT, Sultanpur
ABSTRACT
Software cost estimation is a process to calculate effort, time and cost of a project, and assist in better
decision making about the feasibility or viability of project. Accurate cost prediction is required to
effectively organize the project development tasks and to make economical and strategic planning, project
management. There are several known and unknown factors affect this process, so cost estimation is a very
difficult process. Software size is a very important factor that impacts the process of cost estimation.
Accuracy of cost estimation is directly proportional to the accuracy of the size estimation.
Failure of Software projects has always been an important area of focus for the Software Industry.
Implementation phase is not the only phase for Software projects to fail, instead planning and estimation
steps are the most crucial ones, which lead to their failure. More than 50% of the total projects fail which
go beyond the estimated time and cost. The Standish group‘s CHAOS reports failure rate of 70% for the
software projects. This paper presents the existing algorithms for software estimation and the relevant
concepts of Fuzzy Theory and PSO. Also explains the proposed algorithm with experimental results.
KEYWORDS
Software Cost estimation, Particle swarm optimization, Fuzzy logic etc
1. INTRODUCTION
Software cost prediction is a crucial, essential and an important issue for software engineering
research communities. As the software size varies from small to medium or large, the need for
accuracy or correctness in software cost estimation with understanding has also grown. Software
cost estimation is a method to determine effort, time and cost of a project, which in turn helps in
better decision making about the feasibility and/or viability of the project. To effectively organize
the project development tasks and make considerable economical and strategic planning, project
management requires accurate software cost estimation. Cost estimation is a very difficult process
because several known and unknown factors affect the estimation process. Software size is a very
important (desirable) factor that impacts the process of cost estimation. Accuracy of cost
estimation is directly proportional to the accuracy of the size estimation.
Software estimators sometimes confuse size and effort. Size, in a software development context,
is the complete set of business functionalities that the end user gets when the product is deployed
and is in use, and the Person months required to produce the software application of a given size
is the effort.
Failure of Software projects has always been an important area of concern for the Software
Industry. Implementation phase is not the only phase for Software projects to fail, instead
planning and estimation steps are the most crucial ones, which leads to their failure. More than
50% of the total projects which go beyond the estimated time and cost fail.

114
2. MOTIVATION
Based on literature review and current trends in the software industry, some of the important
motivators for this research work are discussed below:
1.) Insignificant Contribution by the various Algorithmic Techniques
There are multiple studies available which accounts for the incorrect and inaccurate
contribution by the various algorithmic techniques for software cost estimation for
various projects. The errors are far beyond the acceptable limits and hence impacting the
overall project. Since these types of traditional techniques and modeling equations based
on algorithmic approach has failed, it has become imperative for us to look out for better
options either in terms of new methods or optimization of the existing ones. This is quite
motivating for a researcher to explore new options.
2.) Emergence of Soft Computing & Meta-heuristic Techniques
Soft Computing techniques like Fuzzy systems and meta-heuristic algorithms like particle
swarm optimization, firefly algorithm, harmony search, and the likes have enhanced their
presence and utility in the software industry. With so many applications existing for such
techniques, software cost estimation is an area which can be explored through their
applications. Their advantageous characteristics can be imparted on the estimation
techniques which can become more efficient. So it is also an important motivation to
study the application of such techniques in the software cost estimation scenario.
3.) Increasing criticality of Software Cost Estimation techniques & rising impact on
Business Activities
These days, the importance of Software Cost Estimation has gone beyond limits. The criticality of
the software is increasing exponentially and thus cost is becoming a very important parameter. A
small error in the process of cost estimation can destroy the organization‟s reputation and
recognition in the market. With digital world becoming so powerful, this negative publicity can
spread virally. Thus cut-throat competitions are making it very difficult for everyone to survive
and better techniques must be researched for cost estimation. Once the estimation process is
completed successfully, one can easily execute the project without much pressure from the
costing point of view.
3. INTRODUCTION TO SOFTWARE EFFORT ESTIMATION
Estimation of effort and time involved in the development of the software product under
consideration are the most critical activities of the software life cycle. This process may be
performed at any stage during the software development and is termed as software cost
estimation. It is necessary to perform the estimate during early phases of the Software
Development Life Cycle (SDCL), which helps us in taking an early decision regarding feasibility
of the software development project. The results of estimation are also used for early analysis of
the project efforts [1].
The primary components [2] of project development costs are: Hardware costs, Operational costs
and Effort costs (in terms of Man-hours deployed on a particular project with their salaries as per
their time utilized on it). The most important (dominant) cost out of all the three is the effort cost.
It has been found to be very difficult to estimate the effort cost and control it, which can have a
significant effect on overall costs. Software cost estimation is an ongoing process which starts at

115
the beginning of the SDLC at the requirement gathering and proposal stage and continues
throughout the conduction of a project. Projects are bound to have a specific monetary plan in
terms of revenue and expenditure, and consistent cost estimation technique is necessary to ensure
that spending is in line with the budget plan. Man-hours or man-months are the units used to
measure the effort. An ill estimated project or poor feasibility study for a project hampers the
image of the software development organization. If a project cannot be completed within the
estimated time or efforts, the budget may get disturbed. This in turn can bring lots of losses for
both the client and the organization. Thus we need to have a strong estimation model available
with us, which can facilitate us in preparing a correct and feasible plan.
Though there has been various models proposed in the past [1] [6], but still the accuracy levels
are not that satisfactory and leaves a scope of improvement over the previous work. In this paper,
a new algorithm which is combination of Particle swarm optimization and fuzzy theory has been
proposed and discussed for the Software cost, effort and time estimation.
4. BACKGROUND
Software cost estimation methods can be majorly categorized into three types: Algorithmic
method, Expert judgment and Analogy based method [7].Machine learning is also an estimation
method but is not a prominently different one in regular practices and is more categorized under
the Analogy based methodology only. Each technique has its own advantages, disadvantages and
limitations. With projects of large cost magnitude, many cost estimation methods should be used
in parallel. This way one can compare the results produced from these methods and reliability
over one particular methodology gets reduced and hence risk is also minimized [8]. But for
smaller or medium sized projects, applying all the cost models might become costly affair thus
making it an expensive start of the project [8].
5. FUZZY LOGIC
Fuzzy logic is a kind of multi-valued logic, which deals with approximate reasoning rather than
exact and absolute reasoning. It is an approach for computing based on degree of truth rather than
just true or false (1 or 0). There are few terms associated with fuzzy logic which are mentioned as
below:
5.1 Fuzzy Number
A fuzzy number is a quantity whose value is uncertain rather than exact as in case of single
valued numbers. Fuzzy number refers to a connected set of possible values with each value
having its own weight in range 0 to 1. The weight assigned to each value is called the
membership function.
5.2. Membership Function
As per the definition, for a fuzzy set A on the universe of discourse X, membership function is
defined as
µA:X → [0,1] (1)
where, each element of X is mapped to a value between 0 and 1[3].It represents degree of truth as
an extension of valuation. Membership functions are of different shapes [9] such as triangular,
trapezoidal, piecewise linear, Gaussian, bell-shaped, etc.

116
Trapezoidal Function: It is defined by a - lower limit, d- an upper limit, b- lower support limit,
and an upper support limit c, where a < b < c < d. [3]. The Membership function depicting the
Trapezoidal Function is shown below in equation 2.
(2)
Let A be a fuzzy number, A = [a, b, c, d; w] , w is the weight, 0 < w ≤1 .The Trapezoidal
membership of this Fuzzy number A should satisfy the following conditions:
(a) A is a continuous mapping from R to the closed interval in [0, 1].
(b) A(x) = 0, where infinite ≤ x ≤ a and d ≤ x ≤ infinite.
(c) A(x) is monotonically increasing in [a, b].
(d) A(x) = w where b≤ x≤ c.
(e) A(x) is monotonically decreasing in [c, d].
Figure 1 Trapezoidal Membership Function within the range of 0 and 1 [4]
In the Figure 1, the x axis represents the universe of discourse, whereas the degrees of
membership in the [0, 1] interval is represented by y-axis [3].
6. PARTICLE SWARM OPTIMIZATION (PSO)
Particle swarm optimization (PSO) is a stochastic optimization technique invented by Dr.
Eberhart and Dr. Kennedy in 1995[5], used the concept of population and social behaviour of bird
flocking or schooling. PSO is an evolutionary computation technique such as Genetic algorithm.
PSO optimizes problem by iteratively trying to improve a candidate (initial) solution with regard
to a given measure of quality or fitness function. Initially PSO starts with candidate solutions,
known as particles and move these particles around in the search-space according to the
mathematical formulae describing the particle‟s position and velocity. Each particle‟s movement
depends on its local best known position but, is also diverted toward the best known positions in

117
the search space [10], [11]. These local best positions are updated as better positions are found by
other particles.
6.1. PSO Algorithm
The steps of PSO Algorithm [5] are summarized as follows
1. For each particle i = 1, ..., S do:
1.1 Initialize the particle's position with a uniformly distributed random
vector: xi ~ U(blo, bup), where blo and bup are the lower and upper boundaries of the
search-space.
1.2 Initialize the particle's best known position to its initial position: pi ← xi
1.3 If (f(pi) < f(g)) update the swarm's best known position: g ← pi
1.4 Initialize the particle's velocity: vi ~ U(-|bup-blo|, |bup-blo|)
2. Until a termination criterion is met (e.g. number of iterations performed, or a solution
with adequate objective function value is found), repeat:
2.1 For each particle i = 1, ..., S do:
2.1.1 Pick random numbers: rp, rg ~ U(0,1)
2.1.2 For each dimension d = 1, ..., n do:
2.1.2.1 Update the particle's velocity: vi,d ← ω vi,d + φp rp (pi,d-xi,d) +
φg rg (gd-xi,d)
2.1.3 Update the particle's position: xi ← xi + vi
2.1.4 If (f(xi) < f(pi)) do:
2.1.4.1 Update the particle's best known position: pi ← xi
2.1.4.2 If (f(pi) < f(g)) update the swarm's best known position: g ← pi
3. Now g holds the best found solution.
The parameters ω, φp, and φg are selected by the practitioner and control the behaviour
and efficacy of the PSO method.
7. COCOMO MODEL
Developed by Barry Boehm in 1981 [12], Cost Constructive Model or COCOMO is a cost
estimation model based on algorithmic properties. There were more than 60 projects which were
tested and analyzed under this model before it was formally announced. Since the beginning,
there were three levels of this model which were defined by Boehm. They were Basic,
Intermediate and Detailed. For the purpose of this research, Intermediate COCOMO model has
been used.
7.1. Intermediate COCOMO
The basic COCOMO model is based on the relationship:
DE = a*(SIZE)b (3)
where SIZE is measured in thousand delivered source instructions. The constants a, b are
dependent upon the „mode‟ of development of projects, DE is Development Effort and is
measured in man-months.
There were three modes proposed by Boehm- Organic, Semi-detached and Embedded Mode.
Organic was associated with small teams with known development environment, Semi-detached
followed a mixture of experience and fresher with mode lying between organic and embedded
and finally, Embedded was used for real time projects with strict guidelines and very tight
schedule. The development efforts for various modes of software in Intermediate COCOMO are

118
calculated by equations given in table 3.1. Since the basic COCOMO was not that much accurate,
intermediate COCOMO was developed with introduction of Cost Drivers.
The EAF term is the product of 15 Cost Drivers [12] that are presented below. The various
multipliers of the cost drivers can be categorized from Very Low to Extra High as given.
The 15 cost drivers are broadly classified into 4 categories of product, platform, personnel and
project. The meaning and classification of the cost drivers are as follows:
Table 1 Development Efforts for various modes in Intermediate COCOMO
Development Mode Intermediate Effort Equation
Organic DE=EAF*3.2*(SIZE)1.05
Semi-detached DE=EAF*3.0*(SIZE)1.12
Embedded DE=EAF*2.8*(SIZE)1.2
a) Product: RELY - Required software reliability
DATA - Data base size
CPLX - Product complexity
b) Platform: TIME - Execution time
STOR- Main storage constraint
VIRT - Virtual machine volatility
TURN - Computer turnaround time
c) Personnel: ACAP - Analyst capability
AEXP - Applications experience
PCAP - Programmer capability
VEXP - Virtual machine experience
LEXP - Language experience
d) Project: MODP - Modern programming
TOOL - Use of software tools
SCED - Required development schedule
With a dependency on the projects, various multipliers of the cost drivers will differ and thereby
the EAF may be greater than or less than 1, thus affecting the Effort.
8. PROPOSED EFFORT ESTIMATION APPROACH
Analogy based approach [13] operates with one or two past projects selected on the basis of their
similarity to the target project. Based on this approach following algorithm has been developed to
estimate the effort.
ALGORITHM 1:
1. Establish attributes of planned project i.e. select all the attributes that are required for the
estimation.
2. Estimate the values of attributes established in step 1.Since estimates are uncertain so
fuzzy numbers are used to depict the generated values for attributes.
3. Now from the repository of historical projects find the project that closely matches the
attributes of planned project.
4. Similarity distance is calculated by comparing attribute values of planned project to that
of historical project. The most similar project, though closest, is still not „identical‟ to the
considered project. Hence, its „weight‟ (as described in step 6) will be 1.

119
5. At this step, the values of linear coefficients may be computed using the inputs of
„similarity distances‟ from the existing project using PSO algorithm.
6.
The equation will be similar to linear equation, except that the weights will be adjusted in powers
of distance ratios (powered to s/si, where s is the minimum distance for the best project and si for
the target project).
Figure 2 Generalized framework for estimation using Fuzzy numbers and PSO
The proposed model as shown in Figure 2 is considered as a form of EA (Estimation by Analogy)
and comprises of following main stages:
i. Construction of fuzzy number of attributes.
ii. Finding Similarity distance between planed project and historical project.
iii. Deriving project weights according to the distance.
iv. Applying PSO algorithm to find B, as per the equation Y = B*X, where B is the coefficient
vector, Y is the matrix for current project and X is the matrix for historic projects.
Construction of fuzzy number of attributes
The proposed method chooses COCOMO81 Dataset that describes the project using 15 attributes.
Firstly, value of each attribute is replaced by its corresponding Fuzzy number in order to depict
uncertainty. Fuzzy numbers can be constructed by any of the following two methods: [15].

120
a. Expert opinion
b. From data
Expert opinion is a totally subjective technique. It depends on identifying pessimistic, optimistic
and most likely values for each Fuzzy number [16], whereas construction using data is based on
the structure of data only. Therefore based on DATA, and using Fuzzy modelling [14], the
approach develops the membership functions for each attributes. Let the numeric value of
attributes in Fuzzy Number is represented in the following sequence, where trapezoidal
membership has been considered as given in the section 4.
 Planned Project (P): [a1, a2, a3, a4; wa]
and [a1, a2, a3, a4; wa] represents a fuzzy number for one attribute, where 0 ≤ a1 ≤ a2 ≤
a3 ≤ a4 ≤ 1 and all are real numbers. Therefore, planned project will also have 15 sets of
this type representing 15 attributes.
 Historical Project (H): [b1, b2, b3, b4; wb]
and [b1, b2, b3, b4; wb] represents a fuzzy number for one attribute , where
0≤b1≤b2≤b3≤b4≤1 and all are real numbers. Therefore, historical project will have 15
sets of this type representing 15 attributes.
Finding Similarity between planned project and historical project
To calculate the similarity distance between two fuzzy numbers, the method proposed by
Azzehet. Al has been used [14] which combines the concept of geometric distance, Center of
Gravity (COG) and height of generalized fuzzy numbers. The degree of similarity S(P,H)
between two generalized Fuzzy numbers comprises of three elements:
a) Height of Adjustment Ratio (HAR)
b) Geometric Distance (GD)
c) Shape of Adjustment Ratio (SAF)
Equation for S(P,H) is given by the following equation[14]:
 HAR is used to assess the degree of difference in height between two generalized Fuzzy
numbers. Equation for HAR is given by[4]:
HAR = √ (min (wa/wb ,wb/wa)) (4)
where, wa is weight for attribute of P and wb is weight for attribute of H.
 GD is used to measure the geometric distance between two generalized Fuzzy numbers
including the distance between their x-axis centroid. Equation for GD is given by:
GD = 1 *((a1-b1) + (a2-b2) + (a3-b3) + (a4-b4) + (xa- xb) )/5 (5)
 SAF is used to adjust the geometric distance, like, when the 2 Fuzzy numbers are having
different shapes. Equation for SAF is given by:

SAF = 1 + abs (ya-yb) (6)
Equation for S(P,H) is given by the following equation
S(P,H) = HAR*(1-GD)/SAF (7)
(xa, ya) and (xb, yb) presents the centre of gravity of generalized fuzzy numbers P and H
respectively and calculated using equations 8 and 9:

121
(8)
(9)
Substituting the values of HAR, GD, SAF from equations 4, 5 and 6 respectively in equation 7,
will give the similarity distance between one pair of attribute. Similarly, similarity distances for
rest of the 14 attributes are calculated.
Now, Let Si where i varies from 1 to 15, denote similarity distance between ith attribute of the
two projects we are comparing, then collectively it will give the similarity distance between two
projects denoted by S, given by equation 10:
(10)
In a similar way, similarity distance between planned project and each of historical projects are
calculated to get the most similar project.
Ranking the Closest Project
After calculating the aggregated similarity measure S between the planned project and each
individual analogue project, we will sort the similarity results according to their value. The
project with highest similarity to target project is chosen. We choose project with least similarity
distance because it has highest potential to contribute to the final estimate. We assume that the
effort of the closest project is E which will be utilized for Effort adjustment.
Application of PSO algorithm for determination of Coefficients
For deriving the new estimate only using the closest project is not sufficient in many cases as [6]
it may lead to bad estimation accuracy. Therefore Particle Swarm Algorithm is used to ensure that
all projects are considered according to the extent to which they are similar to the test project. The
framework of implementation of PSO model is shown in Figure 3. Here, the potential benefits of
applying PSO to analogy-based software effort estimation models based on Fuzzy Numbers are
evaluated. A suitable linear model is derived from the similarity distances between pairs of
projects for obtaining the weights to be used for PSO. PSO may be considered as a process of
random allocation of coefficient values, and then randomly changing them till they are closest in
linking the projects (minimum error).
Figure 3, presents the iterative process to optimize the coefficients of linear equation and software
cost estimation process that combines the best linear equation with estimation by analogy using
Fuzzy number [15].

122
Figure 3 Framework of Adjusted Fuzzy Analogy based Estimation
The algorithm for PSO implementation goes as follows:
1. Pick 100 swarms corresponding to the coordinates of 15 attributes for the equation e =
Sum (abs((A(i,j)*B(j)) – E(i))^(s/si))/(no. of past projects), where A represents the values
corresponding to the original scaled ratings (1-6), B is the quantity to be measured, and s
is similarity ratings. Thus, dissimilar projects get lower share of contribution.
2. Pick random positions and velocities for 15 attributes for each of the swarms. Random
velocity would be in the range (-x,+x), where x could take a value like 1, and position can
be taken in range (0,y). Later, when equation was modified as e = Sum (abs((B(j)/A(i,j))
– E(i))^(s/si))/(no. of past projects), most of the coefficients were found in the range
(0,10). The first attempt would take y = 10.
3. Compute total Error = Sum (| Estimated – Actual effort|/(Actual effort + Estimated
effort)), for each project.
4. New velocity = unit random velocity * error for the swarm
5. No. of iterations may be varied as per the accuracy requirements.
6. The „particle best‟ and „global best‟ coordinates may be computed as per the standard
PSO method.

123
A representation of the PSO algorithm for the project is shown in Fig. 4.
Figure 4 Adjusting Reused Effort process diagram
The failure of implementation of an iterator may come in several forms, such as
(i) Poor/Slow convergence, or slowdown before convergence near the required point,
(ii) Divergence
(iii) Undesirable oscillations
These problems may be prevented if the iteration function is mathematically consistent. This may
be ensured as follows:
 The swarm movement starts slowing down when the error value start decreases, leading
to a tendency to converge around the best value, as the number of iterations are increased.
 To ensure that the 15 coefficients do not go too high, the normalized data is inverted to
form the equation E = Sum (B/S) + e implies e = E – sum (B/S);
e (total) = sum [ abs({ E – sum(B/S) }) ^ s/si]
If the absolute value of S increases, the error grows to bring it down again. This keep the
values of coefficients represented by „S‟ in a narrow range.
 An important mathematical step in the algorithm was that of „initialization‟. If initial
coefficients are taken in the range (0, 10), then there are good chances that the errors
would be in the range (0, 1). If velocities are allocated in (0, 1) range, a fast convergence
may be expected.

124
9. EXPERIMENTAL RESULTS
Using the effort of most similar project and calculated similarity distances between each attribute,
the coefficients of all effort drivers in OPTIMTOOL (MATLAB (R2011a)) [104] are found that
results in significantly lesser errors. There is no proof on software cost estimation models to
perform consistently accurate within PRED 25% (number of observation for which MRE is less
than 0.25). Applying PSO algorithm, the value of coefficients of effort drivers (shown in Table 3)
is obtained that aim to minimize the MRE [204] of projects. For some of the effort drivers the
coefficient value is very very less. So we can deduce that these drivers does not play important
role in the adjustment of effort. For the proposed PSO algorithm, Figure 5 shows how algorithm
converges with the number of iterations. Two sets of iterations per simulation set were performed
in order to note the variation of precision as well. It was found that the precision also improves
with the number of iterations, since the absolute difference of error of each pair of simulation
decreases with the number of runs.
Figure 5 PSO algorithm stabilizes as the number of iterations is increased
Table 2 Corresponding to Figure 3.6
A B Avg
10 94.514 12.909 100.968
20 83.157 28.896 97.605
30 73.032 5.996 76.030
40 64.894 9.600 69.694
50 57.730 4.567 60.014
60 50.379 4.414 52.586
70 51.989 3.388 53.683
80 47.565 13.645 54.387
90 48.159 2.691 49.505
100 47.819 6.980 51.309

125
Table 3 Best Coefficient Values Obtained through PSO
Cost
Driver
Coefficents of drivers
(Bi)
ACAP -1.74
AEXP 4.51
CPLX 8.84
DATA 0.39
LEXP 5.43
MODP 6.33
PCAP 3.89
RELY 9.60
SCEP 1.86
STOR 0.97
TIME -2.37
TOOL -1.78
TURN 7.75
VEXP 0.51
VIRT 5.87
10. CONCLUSION AND FUTURE WORK
To solve the problem of uncertain, lost values and ambiguous and vague values of attributes
related to project, the concept of fuzzy numbers is used in the present proposed approach. The
fuzzy model is employed in Estimation by Analogy to reduce uncertainty and improving the way
to handle both numerical and categorical data in similarity measurement. Converting each
attribute (real number) in to fuzzy number solve the problem of ambiguous data. Once it gets
over, calculating the similarity distance of the target project with all historical projects results the
most similar project. But the most similar project still has similarity distance with the project
being estimated. Hence, all projects were considered in varying weights, as per their similarity
distance with the proposed project – the lower the distance, the higher the weight.
Combining the concept of Fuzzy logic and PSO algorithm in a model to estimate the software
effort improves the accuracy of estimation techniques. The experiments, which were conducted
applying this model to NASA63 dataset, establish significant improvement under various
accuracy measures. Case Based Reasoning (CBR), traditional COCOMO and estimation analogy
based Fuzzy Model (EA) have been used for the comparison purpose with the proposed
technique. The evaluation parameter used for the comparison is Mean Magnitude of Relative
Error (MMRE).

126
REFERENCES
[1] Martin Shepperd, and Schofield Chris, "Estimating software project effort using analogies." Software
Engineering, IEEE Transactions on 23.11, pp: 736-743, 1997.
[2] Practical Software Engineering, Accessed from http://ksi.cpsc.ucalgary.ca/courses/451-
96/mildred/451/CostEffort.html on September 5, 2013.
[3] Mohammad Azzeh, Daniel Neagu, Peter I. Cowling, “Fuzzy Grey Relational Analysis for Software
Effort Estimation”, Empirical Software Engineering, Vol. 15, No. 1, pp. 60-90, 2010.
[4] The fuzzy descriptive Logic System, http://gaia.isti.cnr.it/straccia/software/fuzzyDL/fuzzyDL.html
[5] R. C. Eberhart and Y. Shi, “Particle swarm optimization: Developments, applications and resources,”
in Proc. IEEE International Conference on Evolutionary Computation, vol. 1, pp. 81–86, 2001.
[6] MATDOCS : http://www.mathworks.in/matlabcentral/fileexchange
[7] Fiona Walkerden, Ross Jeffery, “An empirical Study of Analogy-based Software Effort Estimation”,
Empirical Software Engineering, Kluwer Academic Publishers, Boston, Volume 4, pp. 135-158,
1999.
[8] Giancarlo Attolini , “Guide to Practice Management for Small- and Medium-Sized Practices”,
International Federation of Accountants ,Third Edition, 2012.
[9] Fuzzy logic membership function, http://www.bindichen.co.uk/post/AI/fuzzy-inference-membership
function.html
[10] K. Vinay Kumar, V. Ravi and Mahil Carr, “Software Cost Estimation using Soft Computing
Approaches”, Handbook of Research on Machine Learning Applications and Trends: Algorithms,
Methods, and Techniques, pp. 499-518, IGI Global, 2010.
[11] Y. Shi, R. C. Eberhart , “A modified particle swarm optimizer”, Proceedings of IEEE International
Conference on Evolutionary Computation, pp. 69–73, 1998.
[12] B. Boehm, “Software Engineering Economics”, Prentice Hall, 1981.
[13] Nikolaos Mittas, Marinos Athanasiades, Lefteris Angelis, “Improving Analogy-based Software Cost
Estimation by a Resampling Method”, Information and Software Technology, 2007.
[14] Mohammad Azzeh, Daniel Neagu, Peter I. Cowling, “Analogy-based Software Effort Estimation
Using Fuzzy Numbers”, The journal of Systems and Software 84, 2010.
[15] M. Jorgensen, “Realim in Assessment of Effort Estimation Uncertainty: It Matter How You Ask.”
IEEE Transaction on software Engineering 30, 2004.
[16] Barry Boehm, Chris Abts, “Software Development Cost Estimation Approaches – A Survey”, 650
Harry Road, San Jose, CA 95120: IBM Research, pp.1-45, 1998.

SOFTWARE COST ESTIMATION USING FUZZY NUMBER AND PARTICLE SWARM OPTIMIZATION

Recommended

More Related Content

What's hot (19)

Viewers also liked (12)

Similar to SOFTWARE COST ESTIMATION USING FUZZY NUMBER AND PARTICLE SWARM OPTIMIZATION (20)

Recently uploaded (20)

SOFTWARE COST ESTIMATION USING FUZZY NUMBER AND PARTICLE SWARM OPTIMIZATION