International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp.

3890-3897
© Research India Publications. http://www.ripublication.com

An Efficient Software Effort Estimation by Combining Neural Network and

Optimization Technique

1
Azath. H, 2Dr. P. Amudhavalli, 3Dr. S. Rajalakshmi and 4Dr. M. Marikannan
1
Research Scholar, Karpagam Academy of Higher Education, Coimbatore, India.

Orcid Id: 0000-0002-7734-7745

2
Research Supervisor & Professor, Department of Computer Science and Engineering, Karpagam Academy of Higher Education,
Coimbatore, India.
Orcid Id: 0000-0001-6979-4847
3
Professor & Head, Department of Computer Science and Engineering, Cheran College of Engineering, Karur, India.
Orcid Id: 0000-0003-4768-1029
4
Professor, Department of Computer Science and Engineering, IRTT, Erode, India.
Orcid Id: 0000-0002-6983-1238

Abstract: etc [3]. While evolution takes place in software, its structure
has a primary impact on enhancement intensive locations, for
Estimating software efforts poses high challenge to both
instance, enhancing the implemented functionality [14, 15].
customers and developers. When the estimated effort is not
Software cost and efforts estimation is become a challenge for
precise, the management may be in menace of agreeing
IT industries. There are lots of methods existing for efforts and
systems that may go beyond their budget, substandard with
cost estimation, but people do not know how to use these
poorly developed functions and time consuming more than
methods [22].
deadline. Hence, estimation of effort in an effective manner
has become a significant task in the software development The software quality models must have the ability to identify
process, despite various difficulties ahead in doing so. The crucial elements accurately. The elements that are determined
process also includes multiple steps to validate the minimized as crucial must allow the application for specific certification
error. This paper introduces a novel approach for effort tasks. The certification task may range from inspecting
estimation based on diverse databases. The approach exploits physically to static and dynamic analysis, testing and
neural network to aid in effort estimation with reduced cost automatic formal analysis techniques. The software quality
and failure ratio. The neural network, which is presented here, models also help in defining the reliability of the dispatched
is enhanced by incorporating optimization process. The products. Numerous statistical models have been reported to
optimization process selects weights for the neural network in estimate fault – proneness of program modules in software
such a way that improvement can be made on the classification engineering [4].
model. In our method, the optimization process is carried out
Software project management is responsible for precisely
using Artificial Bee Colony (ABC) algorithm. The
envisaging the work – effort and duration to develop and/or
performance of the proposed model is investigated using
manage a software system. The process is called software cost
parameters such as Mean Absolute Relative Error (MARE)
estimation [17], which is well-known for its significance on
and Mean Magnitude of Relative Error (MMRE). Performance
software project management. In order to make the managerial
comparison with existing models proves the competitive
decision making process a successful phenomenon, estimation
performance of the proposed model over the conventional
of software development effort must be accurate. Since the
models.
development cost is directly proportional to the project
complexity, the early stage of development requires precise
cost estimation [2]. It also acts as an essential component of
INTRODUCTION
infrastructure projects. Increased estimation accuracy helps the
The primary aim of software engineering can be portrayed as project managers to find alternatives and to get rid of
designing tools and techniques that aid in developing high underestimating technical and business solutions. When the
quality, steady and maintainable applications. Several project reaches the final stage, the accuracy of estimated cost
measurements are used by the developers and managers to will be high because of precise and comprehensive
validate and increase the quality of the application [7]. Any information at this stage [1].
software product on openly available source components
The process of software cost estimation by analogy can be said
depends on many business and technical targets such as
as a prominent machine learning methodology. It can also be
improved quality, short and firm development deadlines,
referred as a basic form of case based reasoning. The analogy
limited development expenditures, source code accessibility,
based estimation is performed on the basis of assumption that

3890
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

like software projects it incurs the same project costs. propagation algorithm has ensured better accuracy than other
Nevertheless, categorical variables on cost have to be handled methods.
by improved techniques [11]. Size, efforts to be put forth,
Reddy et al. [6] have introduced a model to estimate the
development duration, adopted technology and ensured quality
software effort. The model was based on artificial neural
are the potential subjects of estimation in software
network and designed in such a way that the network
development. Despite various effort models have been
performance can be adequately improved so as to meet the
introduced, they consider software size as significant
COCOMO model requirements. Multilayer feed forward
parameter, when the development effort has to be considered
neural network has been used for the model and its parameters,
as the most significant subject. For instance, function point is a
which has aided in estimating the software development effort.
measurement model of software size. It exploits logical
Back propagation learning has trained the network by iterative
function terms that make both users and owners understand
comparison of actual effort and network’s prediction, when a
easily. However, the measured size remains constant, because
training set has been given as input. The neural network model
it considers only the functional requirements irrespective of
has outperformed over other model in terms of estimation
the contribution of programming language, design technology
accuracy, when experimentation has been carried out on
and development skills [13].
COCOMO dataset.
The cost estimation process may face the following problems.
Though various research eras have crossed, no attempt has
 Difficulty ahead in mapping cost – affecting factors been made to identify the most established software effort
with output metrics, because these factors exhibit estimation methods to ensure superior accuracy. Moreover, it
nonlinearity has been reported in the literature that there is no consistent
performance by any of M estimation methods that have been
 Difficulties in measuring the metrics because of considered for study. Certain assertions have been made to
incompleteness and imprecision of data acquired at develop ensembles of various estimation methods rather than
initial stage identifying best estimation method. Ekrem et al. [8] have
 Though numerous models prevail for estimation, introduced a method, which has unified nine learners with ten
challenges remain in identifying suitable model for preprocessing techniques so that 9 10 ¼ 90 solo methods can
the current circumstance and be developed. Experimentation has been carried out using 20
datasets and investigation has been done using seven error
 Difficulties in combined usage of algorithmic and measurements. The experimental outcome has revealed top n
non – algorithmic models [12]. (here, n ¼ 13) individual methods, which have exhibited
Nowadays, the community is highly relying on information performance stability on varying datasets and error
technology and software systems. This necessitates an measurements. The obtained best 2, 4, 8, and 13 individual
appropriate estimation method as it has a wide impact on methods have been unified to produce 12 multi – methods that
project cost and quality. Numerous estimation models have have been further subjected to compare with the individual
been reported in the literature, where more researchers sill methods.
intend to formulate better estimation models [18]. The rest of Given a small significant content, there should be a concise
the paper is organized as follows. Section 2 presents a review information and minimal value added to the complex learning
report on recent research works. Section 3 details the proposed schemes. Ekrem et al. [9] have introduced a QUICK method,
methodology, which includes effort estimation using neural which has determined the Euclidean distance between
network and optimization process. Section 4 discusses the instances (rows) and features (columns) of SEE data. Further,
results of the proposed technique. it has pruned synonyms and outliers, which are similar
features and distant instances, respectively, followed by
evaluating the reduced data. The data reduction has been done
RELATED RESEARCHES by comparing predictions that have been obtained from (1) a
The focus on using soft computing techniques for estimating simple learner and (2) a state – of – the – art learner (CART),
software effort has been significantly increased through which used reduced data and collective data, respectively.
various research works. This Section presents the review on Performance investigation has been carried in hold – out
few of the methodologies. experiments using mean and median MRE, MAR, PRED and
MBRE.
Rao et al [5] have estimated cost using a computationally
efficient Functional Link Artificial Neural Network (FLANN). Since software cost estimation process estimates the amount of
By reducing the computational complexity, the neural net can effort and time to put forth to develop a software system, it is
be used for online applications. The network has exhibited a considered as the most significant process. It is also
simple architecture without any full back propagation training considered as a crucial task. A precise estimate can provide a
and hidden layer. The adaptive neural network has worked strong fundamental for the development procedure. Anupama
efficiently by firstly using COCOMO approach for estimating et al [10] have explained Constructive Cost Model
the software cost and then by using FLANN with backward (COCOMO), which is a renowned software cost estimation
propagation. The entire network has remained as “white box” model. Artificial neural network with perceptron learning
because of its comprehensive analysis on every neuron. The algorithm has facilitated the implementation of the model. The
simple architecture and appropriate training by back

3891
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

training and validation of the network have been carried out PROPOSED METHOD FOR SOFTWARE EFFORT
using COCOMO dataset. ESTIMATION
In the recent era, effort to put forth in a software project is the Software designing faces a significant task, called software
most analyzed variable and estimating the effort has been effort estimation, which is a process of estimating the effort to
found as the most tedious task in a project management put forth to construct a software system. However, the process
process. However, a software cost estimation technique of software effort estimation is a tedious task and hence it
facilitates this task by estimating the amount of effort to put remains as a potential research platform since past few
forth and development duration to construct the software decades. The cost and time estimates can be used for initial
system. It has been considered as a significant task to help the coarse validation and for monitoring the progress of project
software sectors in managing the development process development at the development stage, whereas, they can be
proficiently. Numerous cost estimation models have been used for evaluating the productivity of the project at the final
reported in the literature, where each model has portrayed its stage. As a whole, effort estimation aids in developing fulfilled
own merits and demerits in envisaging the development cost proficient software. Our method exploits soft computing for
and effort. Anupama et al. [16] have worked out on estimating estimating the software effort, where classification is
software cost using back propagation neural networks. The performed using neural network. The weights of the neural
developed model can cope up with renowned COCOMO network are optimized using optimization algorithm.
model and work for its performance improvement. It has the
ability to handle imprecision and uncertainty of the input and
to improve the reliability of the estimated software cost. The A.Steps involved in the Effort Estimation Process
experimentation of the model has been carried out on three
openly available software development datasets. Generally, various software parameters are used by the
estimation process to determine the effort to put forth for
The software cost estimation process is a difficult and time developing the particular software. Figure 1 portrays the steps
consuming task. Analogy – based estimation of software effort that constitute the proposed effort estimation process. The
has been claimed as one of the viable techniques in the field. dataset is comprised of numerous parameters that are used to
However, the method remains incapable in a circumstance of determine the actual and estimated efforts followed by
precise handling of categorical data. Initially, the software calculating MRE value. Consequently, these values are
effort estimation models were developed based on regression classified by applying classification process for which neural
analysis and mathematical models. But nowadays, the models network is used. It helps to identify the suitable parameter
are on the basis of simulation, soft computing, genetic value to perform the process. In our method, the neural
algorithm, neural network, fuzzy logic modeling, etc. Ziauddin network is enhanced by optimizing the weights using
et al. [17] have attempted to increase the software effort optimization algorithm. Here, ABC is used to solve the
estimation accuracy by exploiting fuzzy logic model. The purpose. Unifying ABC here helps in improving the
approach has fuzzified the input variables of COCOMO II classification process. Eventually, an effort value from the
model and defuzzified the output variable to obtain the classifier is obtained and it is considered as the near – best
estimated effort. The linguistic terms for COCOMO II model effort value with reduced error rates.
have been defined using triangular fuzzy membership
functions.
Software development effort estimation was considered a
fundamental task for software development life cycle as well
as for managing project cost, time and quality. Therefore,
accurate estimation was a substantial factor in projects success
and reducing the risks. In recent years, software effort
estimation has received a considerable amount of attention
from researchers and became a challenge for software
industry. In the last two decades, many researchers and
practitioners proposed statistical and machine learning-based
models for software effort estimation. Ghatasheh [23] have
proposed Firefly Algorithm as a metaheuristic optimization
method for optimizing the parameters of three COCOMO-
based models. These models include the basic COCOMO
model and other two models proposed in the literature as
extensions of the basic COCOMO model. The developed
estimation models are evaluated using different evaluation
metrics.

Figure 1. Proposed software effort estimation model.

3892
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

The dataset is comprised of numerous parameter values that Employee Bee Phase
have to be subjected to classification. Hence, these values are
There are three bee groups in the artificial bee colony. They
applied to the neural network for classification of relevant
are employed bees, onlooker bees and scout bees. An
parameters. The following section details the classification
employed bee is a bee that flies back to its previously visited
process of neural network.
food source, whereas an onlooker bee is a bee resides in the
dancing area to decide on the food source to be selected. Scout
bees are bees that undergo random wandering for food source.
B. Training in Neural Network
The employed bees contribute the first half of the colony,
The software parameter values are acquired from the dataset whereas the rest is contributed by onlooker bees. Each
and given as input to the neural network classifier to perform employed bee contributes by a food source. In other words, the
classification. The general training of neural network enable number of food sources around the hive and the number of
mapping the input to a specific output. The neural network is employed bees is same.
highly compatible with the classification process. The feed
Initially, the employed bees select random set of food source
forward neural network is exploited in the training phase of
positions for which the nectar amounts are calculated. They
the proposed method. Comparison is made between the
reach the hive and distribute the nectar information of the food
parameter values and the input files followed by classifying
sources to the onlooker bees, which are in the dancing region
the suitable parameters. A neural network is comprised of
of the hive. In the perspective of algorithm, arbitrary set of
three layers namely, input layer, hidden layer and output layer.
The basic architecture of the feed forward neural network can initial population pi with n solutions, where each solution
be observed from Figure 2. is the food source position and S p is the population size is
generated. The solution representation can be given as
hi , where 1  i  n is an N-dimensional vector and N is the
number of parameters to be optimized. Once the population is
initialized, it is subjected to the iterative process that involves
employed bees, onlooker bees and scout bees.

Onlooker Bee Phase

This phase enables the onlooker bees to select the food sources
based on the nectar information obtained from the employed
bees and to generate new set of solutions. Generally, the
onlooker bee is font of food source area, which has substantial
nectar information shared by the employed bees in the dancing
Figure 2. General Feed Forward Neural Network region of the hive. The probability of selecting a food source
Architecture. by an onlooker bee is directly proportional to the nectar
amount of the food source. Hence, the employed bee dancing
with higher nectar value assigns an onlooker bee for the food
ABC, which is an optimization algorithm, works in the position. The probability of selecting a food source ( Fs ) by an
proposed method to allocate weights on different nodes of the onlooker bee can be given as follows.
neural network. In other words, ABC determines relative
weights. fi
Fs  n (1)
 fk
k 1
Proposed Artificial Bee Colony for Optimization of weights
in Neural Network: where,
In an ABC model, a food source position refers to a possible f i refers to fitness of the solution and
solution for the optimization problem, whereas nectar quantity
of a food source refers to the quality (fitness) of the respective n refers to the number of food sources which is equal to the
solution. The objective of bees is set as to determine the best number of employed bees.
solution [20]. Each employed bee shares the information with
an onlooker bee and flies back to the food source, which was The onlooker bee reaches the selected food source and selects
visited by it in the previous wandering. This process is a new neighborhood of the selected food source based on
undertaken since the memory keeps the record of the food visual knowledge. The visual knowledge is obtained by
source. The employed bee selects a new food source using the comparing both the food positions. When the bee abandons a
visual knowledge about the neighborhood of the one stored in food source due to its lesser nectar value, the scout bee
the memory and assesses the nectar amount [19]. generates a new random food source and fills the abandoned
position. The onlooker bee modifies the food source position

3893
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

stored in the memory and hence finds new food source and advantageous at the beginning stages of iterations, but it may
validates the nectar amount (fitness) of it. fail at the final stage of iterations. Hence, global search is
recommended for scout bee at the initial stage and local search
If we consider an old position g i ,k and a new position hi ,k , at the final stage of iterations. As there may not be any
the relationship can be given as improvement even from the best food source at the final
iterations, the scout bees are selected and removed from the
hi,k  g i,k   i,k ( g i,k  g j ,k ), i  j (2) population. Hence, ABC works out to determine the suitable
weight for each network node to increase the classification
Where, performance.
j  {1,2,..., n}
k  {1,2,..., N} RESULT AND DISCUSSION
The proposed method is implemented in Netbeans 7.4, which
 i,k is an arbitrary number in the range [−1, 1]. is a renowned platform to use for effort estimation process
because of its compatibility. The platform offers variety of
The position update equation interprets that a decrease in the
applications that can be developed from a set of components,
deviation between the parameters of g i ,k and g j ,k leads to a which are often referred as modules. As Netbeans has the
decrease in the perturbation on the position g i ,k . Hence collective set of modules as built – in functions to develop
program, it is convenient for any user to initiate the work
adaptive reduction in the step length happens, when optimal immediately. The study considers Desharnais dataset, which
solution in the search space has reached. Reformulating the has 81 projects with incompleteness in 4 projects that can be
position updating step leads to the following equation. removed. There are nine independent variables and one
hi,k  g i,k   i,k ( g i,k  g j ,k ) (3) dependant variable in the dataset.
Despite numerous error measures are in practice, the most
A time domain representation can be given for the position renowned error measure is Mean Absolute Relative Error
update equation by considering g i ,k as Xl when hi ,k is taken (MARE).
as X l 1 . Hence, we obtain

MARE   n ( Ee  Ae / Ae )
e 1
 (6)
X l 1  X l   i,k ( g i,k  g j ,k ) (4)
where,
As X l 1  X l refers to discrete version of the derivative of E - Estimated effort
e
order   1 , we can write
A -Actual effort
D  [ X l 1 ]   i,k ( g i,k  g j ,k ) (5) e

A. Experimental Results
Scout Bee phase
The obtained experimental outcomes from the proposed
The scout bees are the employed bees whose food sources are method are tabulated in Table 1. The actual effort and the
abandoned by the employed and onlooker bees. They search estimated efforts are determined for various sizes followed by
randomly and replace the abandoned food sources by new determining the Magnitude of Relative Error (MRE) for each
food sources. This process can be simulated by replacing the entry. The actual effort is typically lesser than the estimated
abandoned solution by randomly generated solution. A food effort. As per the equation (8) given below in the performance
source is said to be abandoned, when the position does not evaluation section, MMRE for the efforts are determined for
provide any improvements over a pre-defined number of the execution time
iterations, often termed as limits. A typical ABC algorithm
enables the scout bees to search the solutions arbitrarily within
the vicinity of the hive. This style of search may be

3894
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

Table 1. Effort Estimates and MRE

No Original effort Estimated effort MRE

1 7538.342 2199.522 0.708222

2 5037.401 2357.995 0.531902
3 797.177 380.125 0.523161
4 4926.516 1696.964 0.655545
5 3341.558 996.325 0.701838
6 4195.771 1207.623 0.712181
7 3157.767 1095.779 0.652989
8 5510.152 1661.626 0.698443
9 4817.673 3108.069 0.354861
10 2445.469 995.483 0.592928
11 5735.558 1730.208 0.698337
12 7249.34 3546.582 0.510772
13 1452.7 913.559 0.37113
14 5002.478 1745.231 0.651127
15 5674.268 2139.205 0.622999
16 3098.382 775.079 0.749844
17 4038.353 1269.224 0.685708
18 7007.879 1530.824 0.781557
19 7885.784 1900.384 0.759011
20 1633.202 399.754 0.755233
21 9488.004 5987.829 0.368905
22 8337.024 2111.167 0.746772
23 7422.679 2544.948 0.657139
24 7847.99 4252.522 0.458139
25 7045.08 1694.464 0.759483
26 6071.397 1417.001 0.76661
27 5218.878 1544.466 0.704062
28 8758.911 1971.126 0.774958
29 6666.486 2881.585 0.567751
30 7795.007 1842.441 0.763638

B. Performance Analysis The MMRE calculation for the estimated effort can
be done using equation (7). The proposed method has
The equations that are given below are used to determine MRE
accomplished better MMRE than other fuzzy – based works.
and MMRE.
1 n
MRE  ( Ae - Ee ) / Ae (7) MMRE   MRE i (8)
n i 1
where,
where n= number of data * attributes taken
E - Estimated effort
e
A -Actual effort.
e

3895
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

Table 3. MMRE measurements for the proposed and existing

methods.
METHODS MMRE (%)
Proposed Method 0.6428
Fuzzy method 30. 6
Analogy with fuzzy number 26.89

Figure 6 portrays the graphical illustration of comparative

results between the proposed method and the existing method
based on MRE and MMRE measurements. Here, the
Figure 4. Graphical representation for Effort Estimates comparison interms of MMRE is made between the proposed
system and the existing methods given in [21]. The graphical
illustration demonstrates that the proposed method
The graphical illustration of the obtained effort value from the outperforms the existing methods in terms of efficiency.
proposed method is affixed in Figure 4 in which the actual
effort is relatively higher than the estimated effort values.
The accomplished improvement in effort estimation is
observed using MMRE and MARE of the proposed method
prior optimization and post optimization process. The obtained
values are tabulated below.

Table 2. Parameter values before and after applying ABC

algorithm for optimization.
Parameters Proposed Results in (%)
Before Optimization After Optimization
MMRE 13.608 0.6428
Figure 6. The comparison of MMRE measure for the
MARE 65.639 19.28524
proposed method

The performance improvement accomplished from post –

optimization process is better illustrated using the following CONCLUSION
figure, where the error values obtained from prior optimization Designing a software system requires software effort
and post optimization are plotted. estimation significantly. Numerous research works have been
carried out to increase the precision of effort estimate of the
software system. This paper has proposed a novel approach to
estimate the software effort precisely. The approach has been
contributed by neural network classification process and an
optimization process. The neural network has classified
various software parameters. For betterment of classification
performance, ABC has been used to optimize the weights of
neural network. Error parameters such as MRE, MMRE and
MARE have been determined and performance comparison
has been made with the existing method. The experimental
outcomes have demonstrated the proposed system outperform
the existing method in estimating the software effort more
precisely.
Figure 5. Parameter values before and after applying ABC
algorithm for optimization
REFERENCE
The effectiveness of the proposed method is then proved by [1] Jamshid Sodikov, "Cost Estimation of Highway
compariing it with existing methods through MMRE Projects in Developing Countries: Artificial Neural
measurements obtained from proposed and existing methods. Network Approach", Journal of the Eastern Asia
The values are tabulated in Table 2, where MMRE is given in Society for Transportation Studies, Vol. 6, pp.1036-
percentage. 1047, 2005.

3896
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 6 (2018) pp. 3890-3897
© Research India Publications. http://www.ripublication.com

[2] Ch. Satyananda Reddy and KVSVN Raju, "An Optimal Adaptive Neuro Fuzzy use Case Size Point
Neural Network Model for Software Effort Estimation" (ANFUSP)", I.J. Intelligent Systems and Applications,
Int. J. of Software Engineering, IJSE Vol.3, No.1,2010. Vol.6, pp. 14-24,2012
[3] Heikki Orsila, Jaco Geldenhuys, Anna Ruokonen and [14] Parag C. Pendharkar, "Probabilistic estimation of
Imed Hammouda, "Update Propagation Practices in software size and effort", Expert Systems with
Highly Reusable Open Source Components", In. proc.of Applications, Vol. 37, pp.4435–4440,2010.
20th World Computer Congress on Open Source
[15] Andrzej Olszak, Eric Bouwers, Bo Nrregaard Jrgensen
Software, Milano, Italy, Vol. 275, pp.159-170, Sep 7-
and JoostVisser, "Detection of Seed Methods for
10, 2008.
Quantification of Feature Confinement", Inproc.of the
[4] Yue Jiang, Bojan Cukic and Yan Ma, “Techniques for 50th International Conference on Objects, Models,
Evaluating Fault prediction models”, Springer Journal Components, Patterns, 2012.
of Software Engineering, pp. 561–595, 2008.
[16] Anupama Kaushik, A.K. Soni and Rachna Soni, "A
[5] B. Tirimula Rao, B. Sameet, G. KiranSwathi, K. Simple Neural Network Approach to Software Cost
Vikram Gupta, Ch. Ravi Teja and S. Sumana, "A Novel Estimation", Global Journal of Computer Science and
Neural Network Approach for Software Cost Technology Neural & Artificial Intelligence, Vol.13,
Estimation Using Functional Link Artificial Neural No.1, 2013.
Network (FLANN)", International Journal of Computer
[17] Ziauddin, Shahid Kamal, Shafiullah khan and Jamal
Science and Network Security, VOL.9 No.6,2009.
Abdul Nasir, "A Fuzzy Logic Based Software Cost
[6] Ch. Satyananda Reddy and KVSVN Raju, "A Concise Estimation Model", International Journal of Software
Neural Network Model for Estimating Software Effort", Engineering and Its Applications, Vol. 7, No. 2,2013.
International Journal of Recent Trends in Engineering,
[18] Ihtiram Raza Khan and Prof. M. Afshar Alam,
Issue. 1, Vol. 1,2009.
"Software cost estimation using a Neuro-Fuzzy
[7] Jehad Al Dallal, "Mathematical Validation of Object- algorithmic approach", International Journal of
Oriented Class Cohesion Metrics", International Journal Computer Science and Management Research, Vol. 2
of Computers, Vol. 4, No.2, 2010. No.7,2013.
[8] Ekrem Kocaguneli, Tim Menzies, and Jacky W. Keung, [19] Dervis Karaboga, Bahriye Akay, A comparative study
"On the Value of Ensemble Effort Estimation", IEEE of Artificial Bee Colony algorithm, Journal of Applied
transactions on software engineering, Vol. 38, No. 6, Mathematics and Computation, Vol. 214, Pp. 108–132,
2012. 2009.
[9] Ekrem Kocaguneli, Tim Menzies, Jacky Keung, David [20] Dervis Karaboga and Celal Ozturk, Fuzzy clustering
Cok and Ray Madachy, "Active Learning and Effort with artificial bee colony algorithm, Journal of
Estimation:Finding the Essential Content of Software Scientific Research and Essays, Vol. 5, No. 14, pp.
Effort Estimation Data", IEEE Transactions on 1899-1902, 2010.
Software Engineering, Vol.39, No.8,2013.
[21] S.Malathi and Dr.S.Sridhar, "Estimation Of Effort In
[10] Anupama Kaushik, Ashish Chauhan, Deepak Mittal and Software Cost Analysis For Heterogenous Dataset
Sachin Gupta, "COCOMO Estimates Using Neural Using Fuzzy Analogy",International Journal of
Networks", I.J. Intelligent Systems and Applications, Computer Science and Information Security,Vol.10,
Vol.9, pp.22-28, 2012. No.10,2012.
[11] S.Malathi and Dr.S.Sridhar, "Estimation of Effort in [22] Poonam Rijwani, Sonal Jain and Dharmesh Santani,
Software Cost Analysis For Heterogenous Dataset "Software Effort Estimation: A Comparison Based
Using Fuzzy Analogy, “International Journal of Perspective", International Journal of Application or
Computer Science and Information Security,Vol.10, Innovation in Engineering & Management, Vol.3,
No.10, 2012. No.12, 2014.
[12] Divya Kashyap, Ashish Tripathi and Prof. A. K. Misra, [23] Nazeeh Ghatasheh, Hossam Faris, Ibrahim Aljarah and
"Software Development Effort and Cost Estimation: Rizik M. H. Al-Sayyed, "Optimizing Software Effort
Neuro-Fuzzy Model", Journal of Computer Estimation Models Using Firefly Algorithm”, Journal
Engineering, Vol.2, No.4, pp.12-14, 2012. of Software Engineering and Applications, Vol.8,
pp.133-142,2015.
[13] Mohammad Saber Iraji and Homayun Motameni,
"Object Oriented Software Effort Estimate with

3897