World Academy of Science, Engineering and Technology 39 2008

A Comparative Analysis of Fuzzy, Neuro-Fuzzy

and Fuzzy-GA Based Approaches for
Software Reusability Evaluation
Parvinder Singh Sandhu, Dalwinder Singh Salaria, and Hardeep Singh

reusable components from scratch to build and strengthen

Abstract—Software Reusability is primary attribute of software their reusable software reservoir [4]. The cost of developing
quality. There are metrics for identifying the quality of reusable the software from scratch can be saved by identifying and
components but the function that makes use of these metrics to find extracting the reusable components from already developed
reusability of software components is still not clear. These metrics if and existing software systems or legacy systems [6]. But the
identified in the design phase or even in the coding phase can help us issue of how to identify reusable components from existing
to reduce the rework by improving quality of reuse of the component systems has remained relatively unexplored. In both the cases,
and hence improve the productivity due to probabilistic increase in whether we are developing software from scratch or reusing
the reuse level. In this paper, we have devised the framework of code from already developed projects, there is a need of
metrics that uses McCabe’s Cyclometric Complexity Measure for
evaluating the quality of the potentially reusable piece of
Complexity measurement, Regularity Metric, Halstead Software
software. The contribution of metrics to the overall objective
Science Indicator for Volume indication, Reuse Frequency metric
of the software quality is understood and recognized [7]-[9].
and Coupling Metric values of the software component as input
attributes and calculated reusability of the software component. Here,
But how these metrics collectively determine reusability of a
comparative analysis of the fuzzy, Neuro-fuzzy and Fuzzy-GA software component is still at its naïve stage. A neural
approaches is performed to evaluate the reusability of software Network approach could serve as an economical, automatic
components and Fuzzy-GA results outperform the other used tool to generate reusability ranking of software [10]. But,
approaches. The developed reusability model has produced high when one designs with Neural Networks alone, the network is
precision results as expected by the human experts. a black box that needs to be defined, which is a highly
compute-intensive process. One must develop a good sense,
Keywords—Software Reusability, Software Metrics, Neural after extensive experimentation and practice, of the
Networks, Genetic Algorithm, Fuzzy Logic. complexity of the network and the learning algorithm to be
used. Fuzzy systems, on the other hand, require a thorough
I. INTRODUCTION understanding of the fuzzy variables and membership
functions, of the input-output relationships, as well as the
T HE demand for new software applications is currently
increasing at the exponential rate, as is the cost to develop
them. The numbers of qualified and experienced professionals
good judgment to select the fuzzy rules that contribute the
most to the solution of the application. As for the Fuzzy
inference system there is a need of membership rules for fuzzy
required for this extra work are not increasing commensurably categories. It is difficult to deduce these membership rules
[1]. Software professionals have recognized reuse as a with a given set of complex data. Neural nets and fuzzy
powerful means of potentially overcoming the above said systems, although very different, have close relationship: they
software crisis [2], [3] and it promises significant work with impression in a space that is not defined by crisp,
improvements in software productivity and quality [4], [5]. deterministic boundaries [11]. Neural network can be used to
There are two approaches for reuse of code: develop the define fuzzy rules for the fuzzy inference system. A neural
reusable code from scratch or identify and extract the reusable network is good at discovering relationships and pattern in the
code from already developed code. The organization that has data, so neural network can be used to preprocess data in the
experience in developing software, but not yet used the fuzzy system. Furthermore, neural network that can learn new
software reuse concept, there exists extra cost to develop the relationships with new input data can be used to refine fuzzy
rules to create fuzzy adaptive system. Initially treated with
skepticism, the flexibility and power of fuzzy systems is now
Parvinder S. Sandhu is Professor with Computer Science &
Engineering Department, Rayat & Bahra Institute of Engineering & Bio-
well recognized. One major application of fuzzy systems has
Technology, Sahauran, Distt. Mohali, Punjab-140104, India (phone: been in controlling manufacturing processes and various
+91-98555-32004; e-mail: parvinder.sandhu@gmail.com). appliances such as air conditioners and video cameras.
Dalwinder Singh Salaria is Lecturer with Computer Science & Engineering Increasingly fuzzy logic is being combined with other
Department, Lovly Institute of Engineering & Technology, Jalandhar, Punjab-
India.
intelligent system methodologies to develop hybrid fuzzy-
Hardeep Singh is Professor with Computer Science & Engineering expert, neuro-fuzzy, or fuzzy-GA systems. In this paper, we
Department, Guru Nanak Dev University, Amritsar(Punjab)-India.

342
 World Academy of Science, Engineering and Technology 39 2008

have devised the framework of metrics that uses McCabe’s • Calculate the Output of the fuzzy Inference System
Cyclometric Complexity Measure for Complexity corresponding to the Input set
measurement, Regularity Metric, Halstead Software Science • Treat FIS value and the Nearest Match Value as
Indicator for Volume indication, Reuse Frequency metric and Chromosome and convert the values into Binary after
Coupling Metric values of the software component as input multiplying the values with 100
attributes and calculated reusability of the software • Perform the cross over of the Values at a Particular
component. Here, comparative analysis of the fuzzy, Neuro- Point
fuzzy and Fuzzy-GA approaches is performed. • Compare the results

II. METHODOLOGY USED 4. Select the best approach from the step 3 and make
Reusability evaluation System for function Based Software reusability evaluation model for the function based software
Components can be framed using following steps: components.
1. A framework of metrics is proposed for structural analysis
of function-based components [12]. The set of metrics are III. RESULTS AND DISCUSSION
able to target all the essential attributes of function-based The proposed Fuzzy and Neuro-fuzzy based methodology is
software as mentioned by Selby [13] in his latest findings, implemented in MATLAB 7.2. MATLAB (Matrix
so we analyzed, refined and used the following metrics of to Laboratory) environment is one such facility which lends a
explore different structural dimensions of a component: high performance language for technical computing. The
Fuzzy-GA implementation is done in C programming
• Cyclometric Complexity Using Mc Cabe’s Measure [14], language. Fuzzy inference is the process of formulating the
• Regularity Metric[15]
mapping from a given input to an output using fuzzy logic.
• Halstead Software Science Indicator[16]
The mapping then provides a basis from which decisions can
• Reuse Frequency Metric[15]
be made. Following figure gives the snap shot of MATLAB
• Lack of Coupling Metric[12].
window while using FIS editor for 5 inputs and 1 output.
2. Parse the software to generate the Meta information related
to the above Software Metrics. .
3. Evaluate the following approaches for the Modeling of the
reusability data provided in the step 2.
a)Fuzzy Based approach: The steps of the fuzzy based
approach are:
• Deciding the Membership functions
• optimized selection of initial rule-base for the Fuzzy
Inference System (FIS)
• Perform the Inference
• Calculate the % accuracy of the results
b) Neuro-fuzzy based approach: The steps for the Neuro-
Fuzzy based are as follows:
• Build Fuzzy Inference System as discussed in fuzzy
based approach
• Load the Training data NF system with the training
data Fig. 1 Fuzzy Inference System with 5 input and 1 output
• Training of the NF based system is performed using
a hybrid learning algorithm using both least squares Using a given input/output data set, the toolbox function
method and back-propagation. In the forward pass anfis constructs a fuzzy inference system whose membership
the consequent parameters are identified using least function parameters are tuned or adjusted using either a back-
squares and in the backward pass the premise propagation algorithm alone, or in combination with a least
parameters are identified using back-propagation. squares type of method. This allows the fuzzy systems to learn
• After traing pahse the trained NF system is tested from the data they are modeling.
with the testing data
• In the testing phase the results are expressed in terms
of % accuracy
c) Hybrid fuzzy-GA based approach: The steps of the
hybrid fuzzy-GA algorithm are:
• Read the input as the metric values
• Find the nearest match with Example data using
Euclidean Distance

343
 World Academy of Science, Engineering and Technology 39 2008

errors of both fuzzy-logic and neuro-fuzzy are as follows:

Sum of percentage error of fuzzy logic output = 4.13

Average fuzzy logic output error = 4.13/14 = 0.295
Sum of percentage error of neuro-fuzzy output = 3.285
Average neuro-fuzzy logic output error = 3.285/14 = 0.2346

In Table I, Cp, Vol, Cm, Reg, R-F, Reu, Fo and N-f

notations are used for Coupling, Volume, Complexity,
Regularity, Reuse- Frequency, Reusability, Fuzzy logic
Fig. 2 ANFIS model structure output, and Neuro-fuzzy output respectively.
In the Table II, the results of Fuzzy-GA implementation are
Fig. 2 shows the overall ANFIS model structure. This shown. Cp, Vol, Cm, Reg, R-F, Reu, Fo, F-G1 and F-G2
model shows five inputs, three membership functions for each notations are used for Coupling, Volume, Complexity,
input, 56 rules, output membership functions and one output Regularity, Reuse- Frequency, Reusability, Fuzzy logic
i.e. Reusability. output, Fuzzy-GA output1 and Fuzzy-GA output2
respectively.
TABLE I
TRUTH TABLE FOR PERCENTAGE ERROR FOR NF AND FUZZY SYSTEM The overall comparison of the three approaches used is
shown in Table III. After analyzing the table it is deduced that
% %
Cp Vol Cm Reg R-F Reu Fo Nf the hybrid Fuzzy-GA approach produces less % error in
error Error evaluating a software component.

3 1 1 1 1 1 1.5 0.5 1.49 0.49 TABLE II

TRUTH TABLE FOR PERCENTAGE ERROR FOR FUZZY-GA SYSTEM
C V C Re R- Re % F- F-
3 3 3 2 1 2 1.5 0.25 1.98 0.01 Fo
p ol m g F u error G1 G2
3 3 1 2 1 2 1.5 0.25 1.98 0.01
3 1 1 1 1 1 1.5 0.5 2.3 1.48
3 3 2 2 1 2 1.5 0.25 1.98 0.01
3 3 3 2 1 2 1.5 0.25 2.02 1.48
3 1 2 2 1 3 2.49 0.17 2.76 0.08
3 3 1 2 1 2 1.5 0.25 2.02 1.48
2 3 2 2 1 3 1.99 0.336 2.92 0.267

3 3 2 2 1 2 1.5 0.25 2.02 1.48

2 1 1 2 1 3 3.00 0.00 3.10 0.03

3 1 2 2 1 3 2.49 0.17 0.45 2.48

2 2 2 2 1 3 2.5 0.166 3.58 0.193

1 2 2 2 1 4 3.0 0.25 3.85 0.037 2 3 2 2 1 3 1.99 0.336 0.47 1.96

1 2 2 3 2 2 3.51 0.755 5.15 1.575 2 1 1 2 1 3 3.00 0.00 3.00 3.00

1 2 2 3 3 5 4.99 0.002 5.15 0.03 2 2 2 2 1 3 2.5 0.166 0.46 2.48

3 1 2 2 1 3 2.49 0.17 3.1 0.03 1 2 2 2 1 4 3.0 0.25 4.0 3.0

2 3 2 1 3 6 3.51 0.415 4.10 0.317

1 2 2 3 2 2 3.51 0.755 4.58 3.48

1 3 3 2 3 4 1.5 0.625 3.10 0.225

1 2 2 3 3 5 4.99 0.002 5.02 4.96

Table I shows the 15 data values taken from the data 3 1 2 2 1 3 2.49 0.17 0.45 2.48
provided by the different companies to calculate the
percentage error and to discuss about the accuracy of the fuzzy 2 3 2 1 3 6 3.51 0.415 0.9 3.48
logic output and the Neuro-fuzzy output. The fuzzy based
inference system and Neuro-fuzzy system are evaluated on the 1 3 3 2 3 4 1.5 0.625 1.46 1.48
basis of percentage error. The calculations for average output

344
 World Academy of Science, Engineering and Technology 39 2008

TABLE III engineering applications involving control, inference and

COMPARISON ON THE BASIS OF PERCENTAGE ERROR
analysis by and large fuzzy-GA is an efficient technique.
% Error %Error %error

(fuzzy logic output) (Neuro-fuzzy logic output ) (fuzzy-GA)

REFERENCES
0.5 0.49 0.48 [1] E. Smith, A. Al-Yasiri, and M. Merabti, “A Multi-Tiered Classification
Scheme for Component Retrieval,” Proc. Euromicro Conference, 1998,
24th Volume 2, 25-27 Aug. 1998, pp. 882 – 889.
0.25 0.01 0.01
[2] V.R. Basili, “Software Development: A Paradigm for the Future,” Proc.
COMPAC ‘89, Los Alamitos, Calif.: IEEE CS Press, 1989, pp. 471- 485.
0.25 0.01 0.01 [3] B. W. Boehm and R. Ross, “Theory-W Software Project Management:
Principles and Examples,” IEEE Trans. Software Eng., vol.15, no. 7,
1989, pp. 902.
0.25 0.01 0.01 [4] W. Lim, “Effects of Reuse on Quality, Productivity, and Economics,”
IEEE Software, vol. 11, no. 5, Oct. 1994, pp. 23-30.
[5] H. Mili, F. Mili and A. Mili, "Reusing Software: Issues and Research
0.17 0.08 0.01 Directions," IEEE Transactions on Software Engineering, Volume 21,
Issue 6, June 1995, pp. 528 - 562.
[6] G. Caldiera and V. R. Basili, “Identifying and Qualifying Reusable
0.3367 0.267 0.01 Software Components”, IEEE Computer, February 1991, pp. 61-70.
[7] W. Humphrey, Managing the Software Process, SEI Series in Software
Engineering, Addison-Wesley, 1989.
0.00 0.03 0.0 [8] L. Sommerville, Software Engineering, Addision-Wesley, 4th Edition,
1992.
[9] R. S. Pressman, Software Engineering: A Practitioner’s Approach,
0.1667 0.193 0.173
McGraw-Hill Publications, 5th edition, 2005.
[10] G. Boetticher and D. Eichmann, “A Neural Network Paradigm for
0.25 0.0375 0.00 Characterising Reusable Software,” Proceedings of the 1st Australian
Conference on Software Metrics, 18-19 November 1993.
[11] S. V. Kartalopoulos, Understanding Neural Networks and Fuzzy Logic-
0.755 1.575 0.74 Basic Concepts and Applications, IEEE Press, 1996, pp. 153-160.
[12] Parvinder Singh Sandhu and Hardeep Singh, “Automatic Reusability
Appraisal of Software Components using Neuro-Fuzzy Approach”,
0.002 0.03 0.004 International Journal of Information Technology, vol. 3, no. 3, 2006,
pp. 209-214.
[13] Richard W. Selby, “Enabling Reuse-Based Software Development of
0.17 0.03 0.17 Large-Scale Systems”, IEEE Trans. Software Eng., 31(6), (June 2005),
pp. 495-510.
[14] T. MaCabe, “A Software Complexity measure”, IEEE Trans. Software
0.415 0.317 0.42 Eng., vol. SE-2 (December 1976), pp. 308-320.
[15] G. Caldiera and V. R. Basili, Identifying and Qualifying Reusable
Software Components, IEEE Computer, (1991), pp. 61-70.
0.625 0.225 0.63
[16] Maurice H. Halstead, Elements of Software Science (Elsevier North-
Holland, New York, 1977).

IV. CONCLUSION
Fuzzy–GA hybrid algorithm is proved to be best as
compared to the other algorithms considered in this work. In
such data search application the design and developed fuzzy
GA code has shown its superiority because it includes the
advantages of fuzzy as well as genetic algorithms. Fuzzy
provides a robust inference mechanism with no learning and
adaptability while on the other hand, the genetic algorithms
provide an efficient data modification in the wake of
optimization objectives of given application. Neuro-fuzzy
algorithm is definitely superior to fuzzy algorithm as it inherits
adaptability and learning but seriously lacks optimal nature.
From the simulation and the result obtained, it has been shown
that the percentage average error is least in the case of fuzzy-
GA algorithms and maximum in the case of fuzzy algorithms.
Neuro-fuzzy algorithm has yielded accuracy lying between the
accuracy levels as in the case of fuzzy and fuzzy-GA
algorithms. It is concluded that for non linear and complex

345