A Defect Prediction Model for Software Product based on ANFIS

IJSRD - International Journal for Scientific Research & Development| Vol. 3, Issue 10, 2015 | ISSN (online): 2321-0613
All rights reserved by www.ijsrd.com 1024
A Defect Prediction Model for Software Product based on ANFIS
Deepak Kumar Verma1 H. S. Shukla2
1,2
Department of Computer Science
1,2
D.D.U. Gorakhpur University, Gorakhpur, India
Abstract— Artificial intelligence techniques are day by day
getting involvement in all the classification and prediction
based process like environmental monitoring, stock
exchange conditions, biomedical diagnosis, software
engineering etc. However still there are yet to be simplify
the challenges of selecting training criteria for design of
artificial intelligence models used for prediction of results.
This work focus on the defect prediction mechanism
development using software metric data of KC1.We have
taken subtractive clustering approach for generation of fuzzy
inference system (FIS).The FIS rules are generated at
different radius of influence of input attribute vectors and
the developed rules are further modified by ANFIS
technique to obtain the prediction of number of defects in
software project using fuzzy logic system.
Key words: Defect Prediction, ANFIS, Artificial
Intelligence, Fuzzy Logic, ANN
I. INTRODUCTION
Adaptive neuro-fuzzy inference system is a fuzzy inference
system developed in the structure of an adaptive neural
network [1]. By using a hybrid learning procedure, ANFIS
can develop an input-output relation based on both human-
knowledge as fuzzy if-then rules and approximate
membership functions from the stipulated input-output data
pairs for neural network training. This procedure of
developing a FIS using the framework of adaptive neural
network is called an adaptive neuro fuzzy inference system
(ANFIS)[1][2]. There are two techniques employed by
ANFIS to update membership function parameters: 1) for all
the parameters using backpropagation (a steepest descent
method), and 2) a hybrid method made up of back
propagation for the parameters attached with the input
membership functions and least squares estimation for the
parameters linked with the output membership function.
Thus, the training error diminishes, for a smallest amount
locally, all through the learning process. It applies the least-
squares method to identify the consequent parameters that
define the coefficients of each output equation in the
Sugeno-type fuzzy rule base. The training process continues
till the desired number of training steps (epochs) or the
desired root mean square error (RMSE) between the desired
and the generated output is achieved. This study uses a
hybrid learning algorithm, to identify premise and
consequent parameters of first order Takagi-Sugeno type
fuzzy system for predicting software error.
Defective software modules cause software
failures, increase development and maintenance costs, and
decrease customer satisfaction. In ones work one strives to
better software excellence and testing effectiveness by
developing forecasting models from code attribute to allow
an appropriate detection of fault-prone module. To identify
and locate defects in software projects is a difficult task.
Particularly, when project sizes grow up, this task becomes
expensive with sophisticated testing and evaluation
mechanisms. On the other hand measuring of softwares in
an uninterrupted and regimented style brings many merits
such as precise evaluation of project costs and schedules, in
improving qualities of product and process. Detailed
analysis of software metric data also gives important clue
about the locations of possible defects in a programming
code.
In the present work an Adaptive Neuro Fuzzy
Inference System (ANFIS) Approach will be applied for the
development of an efficient predictive model using
Substractive Clustering Algorithm. For this NASA’s Metrics
Data Program (MDP) containing software metric data and
error data at the function/method level has been used to
validate the algorithm [3]. This includes a numeric attribute
(NUMDEFECTS) to indicate defectiveness. The objective
in the construction of models of software error prediction is
to use measures that may be obtained relatively early in the
software development life cycle to provide reasonable initial
estimates of quality of an evolving software system.
II. RELATED WORK
In this section we have discussed recent advancement in the
field of software defect prediction in the last decades. Xiao-
dong Mu et. al.,(2012)[8], in their work to improve the
accuracy of software defect prediction, a coevolutionary
algorithm based on the competitive organization is put
forward for software defect prediction. During this
algorithm, firstly, competition mechanism is introduced to
organization coevolutionary algorithm. Then, three
evolution operators which are reduced operator, allied
operators and disturbed operators are developed for
evolution of population. And competition is considered for
calculate the fitness function. When the algorithm applied
into software defect prediction, it improves the accuracy of
software prediction through increases the diversity of
population.
Manu Banga, (2013) [7], here a new computational
intelligence sequential hybrid architectures involving
Genetic Programming (GP) and Group Method of Data
Handling (GMDH) viz. GPGMDH have been discussed.
Besides GP and GMDH, a host of techniques on the ISBSG
dataset has been tested. The proposed GP- GMDH and
GMDH-GP hybrids outperformed all other stand-alone and
hybrid techniques. It is concluded that the GPGMDH or
GMDH-GP model is the best model among all other
techniques for software cost estimation.
Mohamad Mahdi Askari and Vahid Khatibi
Bardsiri (2014) [4] for the prediction of software defects
used artificial neural network inorder to better the
generalization capability of the algorithm. Further support
vector machine technique was used along with the learning
algorithm and evolutionary technique. Thus this led to the
maximization classification margin and prevented
overfitting problem. This algorithm was tested with eleven
machine learning models from NASA datasets. The
conclusion drawn was that it provided better accuracy and
precision than other models.

(IJSRD/Vol. 3/Issue 10/2015/234)
Mrs.Agasta Adline, Ramachandran. M(2014) [5]
Predicting the fault-proneness of program modules when the
fault labels for modules are unavailable is a challenging task
frequently raised in the software industry. They attempted to
predict the fault–proneness of a program modules when
fault labels for modules are not present. Supervised
techniques like Genetic algorithm based software fault
prediction approach for classification has been proposed.
Xiaoxing Yang, et.al. (2014) [6] Used the rank
performance optimization technique for software forecasting
model development. For this rank to learning approach was
used. The model was developed on previous work and was
later studied for improving the performance of the model. .
The work includes two aspects: one is a novel application of
the learning-to-rank approach to real-world data sets for
software defect prediction, and the other is a comprehensive
evaluation and comparison of the learning-to-rank method
against other algorithms that have been used for predicting
the order of software modules according to the predicted
number of defects. This study shows that the effect of
optimization of the model performance using rank to
learning approach truly improve the prediction accuracy.
III. METHODOLOGY
Fuzzy logic is conceptualized as a generalization of classical
logic. modern fuzzy logic was developed by Lotfi Zadeh [8]
within the mid-1960s to model those issues during which
inaccurate data should be used or in which the rules of
inference are developed during a} very general means
making use of diffuse categories [8]. In fuzzy logic, that is
additionally generally known as diffuse logic, there aren't
simply 2 alternatives however a full continuum of truth
values for logical propositions. A proposition A will have
the truth value zero.4 and its complement Ac the reality
worth zero.5. in line with the kind of negation operator that's
used, the 2 truth values should not be essentially add up to
one. fuzzy logic features a weak association to probability
theory. fuzzy logic doesn't have to be compelled to be even
employing a probabilistic approach. The common route is to
generalize the findings of multivalued logic in such some
way on preserve a part of the algebraical structure [8]. A
fuzzy set theory corresponds to fuzzy logic and therefore the
semantic of fuzzy operators is understood employing a
geometric model. The geometric visualization of fuzzy logic
can provide us a touch on the potential reference to neural
networks. fuzzy logic is used as an interpretation model for
the properties of neural networks, moreover as for giving a
additional precise description of their performance. fuzzy
logic may be used to specify networks directly while not
having to apply a learning algorithmic rule. an expert in a
certain field will generally produce an easy set of control
rules for a dynamical system with less effort than the work
concerned in training a neural network. A classical example
proposed by Zadeh to the neural network community is
developing a system to park a car. it's simple to formulate a
collection of fuzzy rules for this task, however it's not
immediately obvious the way to build a network to try and
do the same nor how to train it. fuzzy logic is currently
being used in several product of business and consumer
electronics for which a decent system is ample and wherever
the question of optimal control doesn't essentially arise.
The process of fuzzy logic is explained in
algorithmic rule 1: foremost, a crisp set of input data ar
gathered and converted to a fuzzy set using fuzzy linguistic
variables, fuzzy linguistic terms and membership functions.
This step is understood as fuzzification. Afterwards, an
inference is created supported a collection of rules. Lastly,
the ensuing fuzzy output is mapped to a crisp output using
the membership functions, within the defuzzification step.
So as to exemplify the usage of a FLS, consider an air
conditioner system controlled by a FLS. The system adjusts
the temperature of the room in line with this temperature of
the room and therefore the target value. The fuzzy engine
sporadically compares the room temperature and therefore
the target temperature, and produces a command to heat or
cool the room.
IV. PROPOSED MODEL
The Proposed model for software defect prediction is based
on Adaptive neuro fuzzy inference system. For this NASA’s
Metrics Data Program (MDP) containing software metric
data and error data at the function/method level has been
used. The process of fuzzy logic is explained in previous
section. For the development of the model a crisp set of
input data are gathered and converted to a fuzzy set using
fuzzy linguistic variables, fuzzy linguistic terms and
membership functions. This step is understood as
fuzzification. Afterwards, an inference is created supported
a collection of rules. Lastly, the ensuing fuzzy output is
mapped to a crisp output using the membership functions,
within the defuzzification step.
A. Algorithm for Proposed Model:
 Step 1: Define Input /Output variables as software
attributes /number of defects.
 Step 2: Select the Significant attributes from all inputs.
 Step3: Generates Fuzzy variables Membership function
for each input and output.
 Step4: Perform clustering of input data at particulat
specified raddi of influence.
 Step 5 : Perform fuzzification of input data.
 Step 6: Define Parameters of Membership function.
 Step 7: Design the fuzzy rules for mapping of output to
input variables.
 Step 8: Evaluates fuzzy output for input data.
 Step 9: Check the MSE of result.
 Step 10: If performance satisfies stopping criteria stop
the generation of FIS system else repeat step 3 to 9.
 Step 11: Generate the modified FIS system using
training data by ANFIS.
 Step 12: Generate output for testing data using ANFIS
based modified FIS system.
In the present work ANFIS Network Structure
model consisting of one input layer with five input variables
and an output layer consisting of weld bead width as the
output variable. This is shown in Fig. 1 below.

Fig. 1: Block structure of Defect Prediction Model using
ANFIS
V. ILLUSTRATION OF PROPOSED MODEL
ANFIS model having twenty input variables are trained and
tested by ANFIS method and their performances compared
and evaluated based on training and testing data. The best fit
model structure is determined according to criteria of
performance evaluation. The performances of the ANFIS
model are shown in Fig. 2 & 3 and their best RMSE values
based on radius of influence r=0.5,0.75 and 1, both for the
r=0.75 training and testing data are 0.01 and 13.25
respectively (Table 1 below).
Training Data 0.014 0.0101 1.844
Testing Data 25.197 13.256 31.372
Overall Data 15.518 8.164 19.2
Table 1: RMSE Values for Datasets after using ANFIS
Fig. 2: RMSE Plot of Training Datasets during ANFIS
Training
Fig. 3: RMSE Plot of Testing Datasets during ANFIS
Training
Further in order to determine the capability and
effectiveness of the model to predict the software error
values MRE has been used. MRE is an indication of the
average deviation of the predicted values from the
corresponding measured data and can provide information
on long term performance of the models; the lower MRE the
better is the long term model prediction. A positive MRE
value indicates the amount of overestimation in the
predicated software error and vice versa.
The MRE of training and testing data sets for
software errors are shown in fig. 4, 5 and 6 below for
different radius of influence values.
Fig. 4: Magnitude of Relative Error for r=0.50
MMRE for diff. radius of influence ( r)
r=0.5 r=0.75 r=1.0
Training Data 0.0181 0.0144 3.396
Testing Data 5.63 2.47 120.53
Table 2: MMRE values corresponding to various radius of
influence
The mean MRE for software error prediction of
training data and testing data for different values of radius of
influence are given in table 8 above. It is an indication of
deviation of the predicted values from the corresponding
measured data and can provide information on long term
performance of the models. The lower deviation, the better
is the long term model prediction. A positive value indicates
the amount of overestimation in the predicted values and
vise-versa. Here also the MMRE value for r=0.75, both for
training and testing datasets are the least.

VI. CONCLUSION
The objective in the construction of models of software error
prediction is to use measures that may be obtained relatively
early in the software development life cycle to provide
reasonable initial estimates of quality of an evolving
software system. For this use of artificial intelligence
technique, viz. ANFIS for the development of software
defect prediction model is a very appropriate technique
because predicting the defective modules in a software
system prior to project deployment is a very crucial activity,
since it leads to a decrease in the total cost of the project and
an increase in overall project success rate. Defect prediction
will give one more chance to the development team to retest
the modules or files for which the defectiveness probability
is high. In the present work it has been observed that the
radius of influence effects the subtractive-clustering
performance and at proper selection of parameters at r=0.75
we have obtained very low prediction error.
Note: For the development and illustration of
proposed model NASA’s Metrics Data Program (MDP)
containing software metric data and error data at the
function/method level has been used.
ANNEXURE-1
A. Procedure of the Work Carried Out:
1) Initiate MATLAB.
2) Load data. For this NASA’s Metrics Data Program
(MDP) containing software metric data and error
data at the function/method level has been used.
3) Divide the data into training and testing datasets
using Matlab commands.
4) Start ANFIS Editor using commands.
5) Load training data into ANFIS editor.
6) Generate Fuzzy Inference System (FIS) using
Substractive clustering algorithm.
 Input Selection:- Number and type of
input / output membership functions.
7) ANFIS Training
 Optimization method selection: Error
tolerence, no. of epochs.
8) ANFIS Testing
 Plot ANFIS output against Observed
training and testing data.
9) Record the plot of training and testing datasets.
10) ANFIS analysis based on RMSE i.e. RMSE <
RMSE ref ?
11) NO, goto step 5 and repeat step 6 to 10, else
12) END.
ANNEXURE-2
A. Program for Defect Prediction Model:
1) GENFIS1
shufflingtrgdata=data(randperm(145),1:22);
a = shufflingtrgdata;
a=data;
A=a(:,1);
B= a(:, 2);
C= a(:, 3);
D= a(:, 4);
E= a(:, 5);
F= a(:, 6);
G= a(:, 7);
H= a(:, 8);
I= a(:, 9);
J= a(:, 10);
K= a(:, 11);
L= a(:, 12);
M= a(:, 13);
N= a(:, 14);
O= a(:, 15);
P= a(:, 16);
Q= a(:, 17);
R= a(:, 18);
S= a(:, 19);
T= a(:, 20);
U= a(:, 21);
V= a(:, 22);
a) >> % Prepare Training Data
trn_data(:, 1) = A(1:90);
trn_data(:, 2) = B(1:90);
trn_data(:, 3) = C(1:90);
trn_data(:, 4) = D(1:90);
trn_data(:, 5) = E(1:90);
trn_data(:, 6) = F(1:90);
trn_data(:, 7) = G(1:90);
trn_data(:, 8) = H(1:90);
trn_data(:, 9) = I(1:90);
trn_data(:, 10) = J(1:90);
trn_data(:, 11) = K(1:90);
trn_data(:, 12) = L(1:90);
trn_data(:, 13) = M(1:90);
trn_data(:, 14) = N(1:90);
trn_data(:, 15) = O(1:90);
trn_data(:, 16) = P(1:90);
trn_data(:, 17) = Q(1:90);
trn_data(:, 18) = R(1:90);
trn_data(:, 19) = S(1:90);
trn_data(:, 20) = T(1:90);
trn_data(:, 21) = U(1:90);
trn_data(:, 22) = V(1:90);
b) >> % Prepare Checking Data
chk_data(:, 1) = A(91:145);
chk_data(:, 2) = B(91:145);
chk_data(:, 3) = C(91:145);
chk_data(:, 4) = D(91:145);
chk_data(:, 5) = E(91:145);
chk_data(:, 6) = F(91:145);
chk_data(:, 7) = G(91:145);
chk_data(:, 8) = H(91:145);
chk_data(:, 9) = I(91:145);
chk_data(:, 10) = J(91:145);
chk_data(:, 11) = K(91:145);
chk_data(:, 12) = L(91:145);
chk_data(:, 13) = M(91:145);
chk_data(:, 14) = N(91:145);
chk_data(:, 15) = O(91:145);
chk_data(:, 16) = P(91:145);
chk_data(:, 17) = Q(91:145);
chk_data(:, 18) = R(91:145);
chk_data(:, 19) = S(91:145);
chk_data(:, 20) = T(91:145);
chk_data(:, 21) = U(91:145);

chk_data(:, 22) = V(91:145);
fismat = genfis1(trn_data,2,’gaussmf’);
fismat = genfis1(trn_data,2,’gbellmf’);
XIN=[trn_data(:,1) trn_data(:,2) trn_data(:,3) trn_data(:,4)
trn_data(:,5) trn_data(:,6) trn_data(:,7) trn_data(:,8)
trn_data(:,21)]
2) XOUT=[ trn_data(:, 22)]
>> fismat = genfis2(XIN, XOUT,0.5); showfis(fismat);
% The initial MFs for training are shown in the plots.
for input_index=1:21,
subplot(3,7,input_index)
[x,y]=plotmf(fismat,'input',input_index);
plot(x,y)
axis([-inf inf 0 1.2]);
xlabel(['Input ' int2str(input_index)]);
end
>>[fismat1, error1, ss, fismat2, error2]=anfis(trn_data,
fismat, 1000, [], chk_data);
% The output MFs for training are shown in the plots.
for input_index=1:21,
subplot(3,7,input_index)
[x,y]=plotmf(fismat1,'input',input_index);
plot(x,y)
axis([-inf inf 0 1.2]);
xlabel(['Input ' int2str(input_index)]);
end
index=1:90
input=[trn_data(:,1:21)];
anfis_output1=evalfis(input, fismat1)
index=91:145
input=[chk_data(:,1:21)];
anfis_output2=evalfis(input, fismat2)
plot(1:55,chk_data(:,22), 1:55,anfis_output2)
legend(‘Actual’,’Predicted’)
plot(1:90,trn_data(:,22), 1:90,anfis_output1)
legend(‘Actual’,’Predicted’)
REFERENCES
[1] “Fuzzy Logic Toolbox”, MATLAB version R2012a.
[2] JANG, J-S. R., “ANFIS-Adaptive-Network Based
Fuzzy Inference System”, IEEE Transactions on
Systems, Man and Cybernatics, 23(3), pp 665-685,
1993.
[3] http://mdp.ivv.nasa.gov/mdp_glossary.html.
[4] Mohamad Mahdi Askari and Vahid Khatibi Bardsiri
(2014), “Software Defect Prediction using a High
Performance Neural Network”, International Journal of
Software Engineering and Its Applications Vol. 8, No.
12 (2014), pp. 177-188. 14.
[5] Mrs.Agasta Adline, Ramachandran. M(2014),
“Predicting the Software Fault Using the Method of
Genetic Algorithm”, International Journal of Advanced
Research in Electrical, Electronics and Instrumentation
Engineering, Vol. 3, Special Issue 2,, pp 390-398.
[6] Xiaoxing Yang, et.al. (2014), IEEE TRANSACTIONS
ON RELIABILITY, This article has been accepted for
inclusion in a future issue of this journal.
[7] Manu Banga, “Computational Hybrids Towards
Software Defect Predictions”, International Journal of
Scientific Engineering and Technology Volume 2 Issue
5, pp: 311-316, 2013.
[8] Xiao-dong Mu, Rui-hua Chang, Li Zhang, “Software
Defect Prediction Based on Competitive Organization
CoEvolutionary Algorithm”, Journal of Convergence
Information Technology (JCIT) Volume7, Number5,
2012.

A Defect Prediction Model for Software Product based on ANFIS

More Related Content

What's hot (16)

Viewers also liked (14)

Similar to A Defect Prediction Model for Software Product based on ANFIS (20)

More from IJSRD (20)

Recently uploaded (20)

A Defect Prediction Model for Software Product based on ANFIS