See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/326074736

A Literature Review Study of Software Defect Prediction using Machine

Learning Techniques

Article · June 2018

DOI: 10.23956/ijermt.v6i6.286

CITATIONS READS

14 2,052

3 authors, including:

Feidu Akmel Ermiyas Birihanu

Wolkite University, Ethiopia Wolkite University, Ethiopia
4 PUBLICATIONS   37 CITATIONS    13 PUBLICATIONS   47 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

mailatgarima@yahoo.co.in View project

All content following this page was uploaded by Feidu Akmel on 09 September 2019.

The user has requested enhancement of the downloaded file.

 International Journal of
Emerging Research in Management &Technology Research Article June
ISSN: 2278-9359 (Volume-6, Issue-6) 2017

A Literature Review Study of Software Defect Prediction

using Machine Learning Techniques
Feidu Akmel* Ermiyas Birihanu Bahir Siraj
(MSc.) Computer Science, (MSc.) Software Engineering, (MSc. Candidate) Addis Ababa
Wolkite University, Wolkite, Wolkite University, Wolkite, Science and Technology University,
Ethiopia Ethiopia Ethiopia

Abstract-

S
oftware systems are any software product or applications that support business domains such as
Manufacturing,Aviation, Health care, insurance and so on.Software quality is a means of measuring how
software is designed and how well the software conforms to that design. Some of the variables that we are
looking for software quality are Correctness, Product quality, Scalability, Completeness and Absence of bugs,
However the quality standard that was used from one organization is different from other for this reason it is better to
apply the software metrics to measure the quality of software. Attributes that we gathered from source code through
software metrics can be an input for software defect predictor. Software defect are an error that are introduced by
software developer and stakeholders. Finally, in this study we discovered the application of machine learning on
software defect that we gathered from the previous research works.

Keywords: Machine Learning, Software Defect, Software Engineering, Machine Learning Techniques

I. INTRODUCTION
When there repeatedly exists a software failure in system through time it automatically leads to software defect. Software
defect are an error that are introduced by software developer and stakeholders. The main objective of software defect
prediction is to improve the quality, minimized cost and time of software products. Software defect is also referred to as
bug can be defined as shortage in the software product that causes the software not to perform its task as the programmer
and customer needed.
Machine Learning is one of the most vital and motivating area of research with the objective of finding meaningful
information from huge data sets. The basic purpose of machine learning is to extract useful pattern from the data, mining
data may be structured format (example. multiple data base) or text mining: unstructured data (example, natural language
document).
In this study, we deeply observed the major factor of software failures that lead the software company to software defect,
consume cost and times to test and maintenance after delivered to the stakeholders. In addition, we examine the
recommended solutions to software failures, machine learning concepts and application of machine learning on software
engineering, specifically for software testing and maintenances.
The rest of this paper is organized as follows:Section 2 explains major factor for software failures and the
recommendations filter out by the researcher;Section 3 summarized common Known defect predictors; Section 4
discussed about machine learning concepts and the application area of machine learning with regarding of software
engineering; specifically, for software defects.Finally, we tried to assess the research works with respect to machine
learning on software defect and classified it based on methods they used namely based on classification method,
Clustering method and ensemble methods. The researchers conclude this paper with a summery and provide the future
direction for researchers.

II. MAJOR FACTOR OF SOFTWARE FAILURE

Software systems are any software product or applications that support business domains [1], such as manufacturing,
banking, healthcare, insurance, aviation, social networking, e-commerce or any other domains. In order to develop and
design software system, there is a need of finance, human experts in domain area, a lot of time, tools and infrastructures.
Now a day, even if the software company has a lot of experience in designing and developing projects, (according to
Table 2) the software failure is increasing from time to time, which leads to loss of capital, time and energy consumption.
The system may fail due to fault happened every software development cycle or customer may not provide you the exact
requirement because they do not have awareness of information technology projects or political, cultural issues.
The survey participants were also asked about the factors that cause projects to be challenged. The most common known
cause of software failures is Lack of user involvement[1][2][3],unclear goals and objectives[3][4], incomplete
requirement specification[3][4], lack of resources[3], inadequate project planning and scheduling [3-5], Poor
communication between Team members[3][4] and Poor Testing. From Table-1 below, we have been looking that lack of
user input and incomplete user requirement specification are the major factors for projects to be successful. The major
factors for software failure are depicted in Table I below.

© www.ermt.net All Rights Reserved Page | 300

 Akmel et al., International Journal of Emerging Research in Management &Technology
ISSN: 2278-9359 (Volume-6, Issue-6)
Table I Major factor for software failures[6]

According to CHAOS MANIFESTO in the 2013 report, they were doing the survey from 2004 to 2012 software projects
and categorizing it based on the percentage of failure, success and challenging projects as shown in Table II below.

Table II Percentage detail from 2004 – 2012[7]

2004 2006 2008 2010 2012
Successful 29% 35% 32% 37% 39%
Failed 18% 19% 24% 21% 18%
Challenged 53% 46% 44% 42% 43%

A. Recommendation to Address Software Failures

As we have discussed section II, we mentioned the cause of software failures, however, any trouble has a solution here
we list out some of the recommendations for success the projects.
 Monitor the project - the accuracy of the estimate.
 Fully satisfied the user requirements.
 Do not depend on a single cost or schedule estimate.
 Put the clear objectives and goal.

III. SOFTWARE DEFECT PREDICTOR

The software defect predictor is the way or a method that helps for software testing and software development life cycles.
Software defect is also referred to as bug can be defined as shortage in the software product that causes the software not
to perform its task as the programmer and customer needed. There are different types of classification for software
defects, according to the Institute of Electrical and Electronics Engineers (IEEE) standard 104[8] which are:
 Defect: The lack to perform the tasks and system requirements that are provided by developer and customers.
 Error: This can be happening by customer due to the knowledge about the deficiency the software, produce
incorrect results
 Failure: Stop its previous required function of the software products or incorrect result will be provided for
each input that is given by the customers.
 Fault: it is clear or obvious error occurred in the software products.

Depending on the above software defect classifications, a defect that is occurring in software product lead malfunctioning
and health problems in case of serious and critical software, such as NASA software products for Space sciences.

IV. MACHINE LEARNING CONCEPT

One of the main differences between how people and computer work is that human, performing any kind of activity
usually simultaneously expand efforts to improve the way they perform it. This is to say, human can perform any tasks
with the versatility means while not machines.
Machine learning algorithms ‘learn’ to predict outputs based on previous examples of relationships between input data
and outputs. The models that are constructed based on the relationship between input and output slowly improved by
testing and its predictions and correcting when wrong[9]; according to Tom Mitchell definition, the machine learns with
respect to particular tasks T, performance P and Experience E[10].

© www.ermt.net All Rights Reserved Page | 301

 Akmel et al., International Journal of Emerging Research in Management &Technology
ISSN: 2278-9359 (Volume-6, Issue-6)
A. Machine Learning Tasks
A system that is used to understand the concept and its environment using a simplified interpretation of the environment
using model is called cognitive system [11]. The step that we pass to construct the model is known as inductive learning.
The Cognitive system is able to combine its experience by constructing new structures is patterns. The constructed model
and pattern by cognitive system is called machine learning.
Models that are described as predictive since it can be used to predict the output of a function (target function) for a given
value in the function’s domain while informative pattern are characterized only describes the portion of data.
Machine learning task classified into four that are supervised, unsupervised, semi-supervised and reinforcement
learning;however, the most known task is supervised and un-supervised learning [8][12][12]. In section a and section b
below, we looked some of supervised and un-supervised learning below.
i. Supervised Learning
It is the machine learning task of inferring a function from labeled training data. The training data consists of a set of
training examples. In supervised learning, each example is a pair consisting of an input object and a desired output value.
It has two known supervised learning tasks, classification and regression. Classification concerns on building predictive
model for function with discrete range, while regression concern on continuous range model building. A lot of
researchers in machine learning are focused on supervised learning[13].
The most common supervised machine learning methods includes concept learning, classification, rule learning, instance
based learning, Bayesian learning, liner regression, neural network and support vector machine[8][11][14].
ii. Unsupervised Learning
This is also called learning from observation. In unsupervised learning the system has to explore any patterns based only
on the common properties of the example without knowing how many or even if there are any patterns. The most
common method in unsupervised learning’s is association rule mining, sequential pattern mining and clustering.

B. Machine Learning in Software Engineering

The Machine learning is not a difficult science [15]. Machine learns automatically from training data and it generate
summaries of data or existing systems in a smaller form while the software engineer could able to use machines in order
to minimize the system development phases time and cost consumptions.
One thing that overcomes machine learning integrating with software engineering is developing machine learning based
solution to software engineering problems [11]. The data and the complexity of pattern in software engineering needs
pre-processed before applying machine learning method, this is similar with other applications.
Component-based approach fault prevention methods have tremendous attentions from developers now a time, which
reduce the development cost and time as well as to improve the quality and reliability of the projects by breaking the
project in to separate components, however it is only practical for finding the problems in the quality of individual
components [16]. One of the solution to resolve the problem is through machine learning approach to software
engineering.
In order to estimate the quality of software we used an attribute such as software development effort, software reliability
and productivity of programmer to predict the important of model in software engineering. Early software quality
prediction to enhance the performance of system via machine learning techniques (case based reasoning and fuzzy logic)
was studied [17].
According to researches, Software engineering problems that are ready to resolved by machine learning, which are
software measurement selection, defect prediction models, software reuse qualification, software requirements gathering,
software quality estimation, project management, software testing[9][18][19][20][21], we also called it an application of
machine learning in software engineering. Among software engineering problems that we listed, researcher chooses
software defect prediction via machine learning methods to conduct this study.
In Table III we were looking machine learning approaches (methods) for software engineering tasks.

Table III SE (software engineering) tasks and applicable ML methods[9]

© www.ermt.net All Rights Reserved Page | 302

 Akmel et al., International Journal of Emerging Research in Management &Technology
ISSN: 2278-9359 (Volume-6, Issue-6)
C. Machine Learning in Software Defect Prediction
The field of machine learning has been growing rapidly, producing a variety of learning algorithms for different
applications. As we discussed in section 2.5.4.1 one of the application of machine learning is software engineering. The
ultimate value of those algorithms is to a great extent judged by their success in solving real-world problems. Therefore,
algorithm reproduction and application to new tasks are crucial to the progress of the field. However, various machine
learning researchers currently publish for software fault prediction model development. Now, we categorizing successful
software defect model into three that are based on classification, clustering and ensemble methods.

V. BASED ON CLASSIFICATION METHODS

EzgiErturk et al. [22] the data set for the experiment are collected from the PROMISE Software Engineering Repository
and applied McCabe software metrics. The algorithm they are using form experiment was SVM, ANN and ANFIS (new
adaptive model proposed), the performance measure was 0.7795, 0.8685, and 0.8573 respectively.
Another successful paper was published by Surndha Naidu et al.[23], the primary goal of this paper was finding the total
number of defects in order to minimize time and cost. The defect was classified into five parameters such as Volume,
Program length, Difficulty, Effort and Time Estimator. They were using ID3 classification algorithm, to classify defects.
Malkit Singh et al. [52] Explored the fault prone of software early software testing were by developing a model with
Levenberg-Marquardt (LM) algorithm based neural network tool for data collected from the PROMISE repository of
empirical software engineering data then comparing the accuracy of LM with the polynomial function-based neural
network. The experiment showed that Levenberg-Marquardt (LM) have higher accuracy (88.1%). So, that neural network
based machine learning has good accuracy.
Saiqa Aleem et al.[24], in this study they were used around fifteen data sets (AR1, AR6, CM1, KC1, KC3 etc) with
various machine learning methods. Measured the performance of each method and finally conclude that SVM, MLP and
bagging had high accuracy and performances.
According to Venkata U.B et al.[2] They evaluate different predictive model for real-time software defect data Sets. The
experiment showed that there are not any exact techniques exist for each data set; however, IBL and 1 R were relatively
better consistency in prediction accuracy compared to with other methods.
According to Martin Shepperd et al. [25] investigation, in order to predict and evaluate software defect, they were using a
novel benchmark framework. In the evaluation stage, different learning schemes are evaluated according to that scheme
selected. Then, in the prediction stage, the best learning scheme is used to build a predictor with all historical data and the
predictor is finally used to predict defect in the new data.

VI. BASED ON CLUSTERING METHODS

Xi Tan et al.[26] experiments the software defect prediction model based on the function cluster for the purpose of
improving the performance of the model. After applying this method, the researcher upgrades the performance 31.6 % to
99.2% recall and precision from 73.8 % to 91.6%.
Jaspreet Kaur et al.[27]investigated that, the fault proneness of object oriented programming via applying k-mean based
clustering approach and finally they conclude as they have got 62.4% of accuracy.
Model building using clustering algorithms (EM and X-means) from three promise repository data (AR3, AR4, AR5)
with an objective of predicting software faults. The first thing in experiment setting was normalizing the data set in to 0
to 1, then attribute selection algorithm applied that was CfsSubsetEval and without attribute reduction. Experiment result
showed that X-means have accuracy (90.48) than other model for AR3 without attribute reduction[27] .

VII. BASED ON ENSEMBLE APPROACHES

Model building using ensemble approach were conducted by Shanthini et al., The purpose of this research was to address
software fault prediction using ensemble approach. The data set was categorized in to three which are method level, class
level and package levels. They were using NASA KC1 data for both method and class level metrics and eclipse data for
package level with an ensemble methods (bagging, boosting, staking and voting). The experiment result shows that
bagging perform better for method and package level data. Method level result using AUC- curve performance
measurement was bagging (0.809), boosting (0.782), staking (0.79) and voting (0.63). Similarly, the performance
measure of package level data using AUC-Curve was bagging (0.82), boosting (0.78), staking (0.72), and voting (0.76).
In case of class level metric, the performance result could not similar to other metrics using AUC-Curve bagging (0.78),
boosting (0.74), staking (0.8) and voting (0.82) [28].
According to Arvinder Kaur et al., the main purpose of the research was evaluating application of random forest for
predicting fault prone class using open source software. The researcher used JEdit open source software with object
oriented metrics to conduct studies. Based on the experiment result the accuracy RF is 74.24 % and its precision is 72 %,
its recall is 79 %, its F-measure is 75 %, and its AUC is 0.81[29].
YI PENG et al., The goal of the paper was to assess the quality of ensemble approaches in software fault prediction with
analytical hierarchal process. The researcher uses 13 different performance measures for 10 publicly NASA MDP data.
An ensemble method used in this paper was Bagging, Boosting and Staking Based on the performance measure
AdaBoost of decision tree gives best result accuracy 92.53 %, in this case decision tree is base classifier[30].
For more clarity, we summarized the above works with objective, methodology adopted, Contribution of studies and
provide over all a remark for studies as Table IV below.

© www.ermt.net All Rights Reserved Page | 303

 Akmel et al., International Journal of Emerging Research in Management &Technology
ISSN: 2278-9359 (Volume-6, Issue-6)
Table IV: Summary of Related Works
Author Objective of the Methodology/approaches/ Key findings Remark
(year) study tools/techniques used
Based on Classification Approaches
EzgiErtu - To predict - SVM, ANN and ANFS - Performance measure - They used NASA
rk et al software fault - 10-fold cross validation using SVM, ANN and MDP and maximum
(2014) test mode ANFS were 0.7795, accuracy was 0.857
- WEKA tool used 0.8685, and 0.8573 for data using ANFS
method.
Malkit - Software defects - Levenberg-Marquardt - Accuracy with LM based - They used Ant 1.7
Singh et prediction at an (LM) algorithm based ANN was 88.1 % data set and
al (2013) early ANN. - Accuracy with PF based on maximum accuracy
Stage of the And ANN was 78.8 % they scored was 88.7
software - Polynomial function (PF) % using LM based
development life based on ANN ANN.
cycle. - MATLAB R2011a.
Saiqa - Comparative - 10-cross validation test - The mean accuracy for - They used NASA
Aleem et machine learning mode Software fault prediction MDP data set and the
al. methods for - MLP model for the given 15 data maximum accuracy
(2015) publically - SVM - SVM (89.29 %) was 99.52 for PC2
available data - Naïve Bayesian - Bagging (89.38 %) data using MLP
using software - AdaBoost - Random forest (89.08 %) method.
prediction model. - Bagging
- Decision Tree
- Random forest
- KNN
Venkata - To predict - 70 % training and 30 % - IBL and 1-R better - They used NASA
U.B et software fault testing consistency accuracy when MDP data and the
al. - Decision Tree compared with other maximum accuracy
(2005) - Naïve Bayes methods was 0.0543 using
- Logistic Regression - IBL ~ MAE (0.0543) MAE performance
- Nearest Neighbor - 1-R ~ MAE (0.0351) on CM1 data
- 1-Rule
- Neural Network
- WEKA tool used
- NASA MDP (CM1, JM1,
KC1 and PC1)
Author Objective of the Methodology/approaches/t Key findings Remark
(year) study ools/techniques used
Martin - Factors for - Meta-analysis for the - They conclude that defect - It was survey on
Shepperd performance of previous studies on prediction researches factor that affects the
et al. software fault software fault prediction should conduct on blind performance of
(2014) prediction such as all relevant and analysis, improve reporting software fault
accuracy high quality primary protocols and inter group prediction accuracy.
studies. protocols.
Xi Tan et - To improve the - Eclipse 3.0 data - Cluster based defect - They used Eclipse
al. performance of - 90 to 10 % data split prediction better than class 3.0 data and the
(2011) software fault based model in both recall maximum accuracy
prediction model and precision 99.2 % (Recall).
using recall and - Recall (31.6% to 99.2%)
precision - Precision (73.8% to
91.6%)
Qinbao - They proposed - NASA MDP and AR Data - The maximum accuracy
Song et and evaluate a sets scored on PC1 data set - They used NASA
al. general frame - 90 to 10 % data split using Naïve Bayesian MDP and Art 1.7;
(2011) work to evaluate - Two data preprocessors method (89.7%) the maximum
software fault - Two attribute selections accuracy was 89.7
prediction model - Naive Bayes for PC1 data using
- J48 Naïve Bayesian.
- OneR

© www.ermt.net All Rights Reserved Page | 304

 Akmel et al., International Journal of Emerging Research in Management &Technology
ISSN: 2278-9359 (Volume-6, Issue-6)
Based on Clustering Approaches
Jaspreet - Predicting fault - KC1 data set used - 62.4 % accuracy scored - They used only KC1
Kaur et proneness of - WEKA data set and
al. object oriented - K-mean maximum accuracy
(2011) programming was 62.4 using K-
using K-mean mean method.
Mikyeon - To predicting - Three promise repository - Xmeans have higher - They used AR3,
g Park et software faults data (AR3, AR4 and AR5) accuracy (90.48) for AR3 AR4 and AR5; the
al. - EM without attribute reduction. maximum accuracy
(2014) - X-Means was 90.48 for AR3
using Xmeans
unsupervised.
Based on Ensemble Approaches
- software fault - NASA KC1 data for both - Bagging perform better - They used KC1
Shanthin prediction using method and class level when compared with other method and class
i. A et al. ensemble metrics approaches in case of level data; the
(2013) approach - eclipse data for package method and package level maximum accuracy
level - Bagging (AUC [0.809]) was AUC-0.809
- They used WEKA - Boosting (AUC [0.782]) using bagging
- Bagging - Staking (AUC [0.79]) method.
- Boosting - Voting (AUC[0.63])
- Staking - But not in class level
- Voting metrics
- Bagging (AUC[0.78])
- Boosting (AUC[0.74])
- Staking ([0.8])
- Voting ([0.82])
Author Objective of the Methodology/approaches/t Key findings Remark
(year) study ools/techniques used
Arvinder - Evaluating - JEdit open source software - Accuracy of RF was 74.24 - They used JEdit
Kaur et application of data % open source and
al. random forest for - OOP metrics - Precision of RF was 72 % maximum accuracy
(2008) predicting faults ,its recall was 79 % and its was 74.24 % using
F-measure was 75 % RF method.
YI - assess the quality - They used 13 different - The maximum accuracy - They used 10 NASA
PENG et of ensemble performance measure scored on AdaBoost (92 MDP data and
al. approaches in - 10 public NASA MDP .53%) maximum overall
(2011) software fault - 10-fold cross validation accuracy scored was
prediction - Bagging 92.53 using
- Boosting AdaBoost
- Staking
- WEKA

VIII. CONCLUSION AND FUTURE WORK

Now a day the development of software based system are increasing from the previous years due to its benefit. However,
the quality of the system is required before it is delivered to end users. In order to enhance the software quality, we have
various quality metrics such as software testing, CMM and ISO standards. Currently software testing becomes more and
more important in the software reliability. Software defects prediction can effectively improve the efficiency of software
testing and guide the allocation of resources. For the error-prone modules, we should spend more resource and time
The main objective of this study was to assess the previous research works with respect to software defect which applies
machine learning method, data set used, tools they used, methodologies, their contribution to science and we classified it
in to three such as based on classification, Clustering and ensemble methods. Finally, it is possible to extend this study by
systematic literature review which includes books, dissertation, tutorial, Thesis.

REFERENCES
[1] Sandeep D and R. S, "Case Studies of Most Common and Severe Types of Software System Failure "
International Journal of Advanced Research in Computer Science and Software Engineering vol. 2, pp. 341-347
August 2012
[2] Venkata U and R. A, "Empirical Assessment of Machine Learning based Software Defect Prediction
Techniques " Proceedings of the 10th IEEE International Workshop on Object-Oriented Real-Time Dependable
Systems 2005.
[3] Robert N, "Why Software Fails " 2005

© www.ermt.net All Rights Reserved Page | 305

 Akmel et al., International Journal of Emerging Research in Management &Technology
ISSN: 2278-9359 (Volume-6, Issue-6)
[4] Rajkumar G and K.Alagarsamy, "The Most Common Factors For The Failure Of Software Development
Project," vol. 11, pp. 74-77, January 2013.
[5] L. J, "Major Causes of Software Project Failures " CROSSTALK The Journal of Defense Software Engineering
pp. 9-12, July 1998.
[6] T. Clancy, "The Standish Group Report CHAOS," Project Smart pp. 1-16, 2014.
[7] Vikas S and J. R, "Cataloguing Most Severe Causes that lead Software Projects to Fail," International Journal
on Recent and Innovation Trends in Computing and Communication vol. 2, pp. 1143– 1147, May 2014.
[8] Mikyeong P and E. H, "Software Fault Prediction Model using Clustering Algorithms Determining the Number
of Clusters Automatically," International Journal of Software Engineering and Its Applications, vol. 8, pp. 199-
204, 2014.
[9] Harry A, "machine learning capabilities, limitation and implications " Technology Futures Researcher at Nesta
2015.
[10] T. M. Mitchell, "Machine Learning " 1997.
[11] George T, Ioannis K, Ioannis P, and Ioannis V, "Modern Applications of Machine Learning " Proceedings of the
1st Annual SEERC Doctoral Student Conference vol. 1, 2006.
[12] Sarwesh S and S. K, "A Review of Ensemble Technique for Improving Majority Voting for Classifier,"
International Journal of Advanced Research in Computer Science and Software Engineering, vol. 1, pp. 177-
180, January 2013.
[13] Christian M, Gail K, and Marta A, "Title," unpublished|.
[14] D. Zhang, "Title," unpublished|.
[15] T. M. Mitchell, The Discipline of Machine Learning: Carnegie Mellon University, 2006.
[16] Xia C, Michael R, Kam-Fai W, and M. W, "ComPARE: A Generic Quality Assessment Environment for
Component-Based Software Systems," Center of Innovation and Technology the Chinese University of Hong
Kong, pp. 1-25.
[17] Ekbal R, Srikanta P, and V. B, "A Survey in the Area of Machine Learning and Its Application for Software
Quality Prediction," ACM SIGSOFT Software Engineering,, vol. 37, September 2012.
[18] L. C. Briand, "Novel Applications of Machine Learning in Software Testing," pp. 3-10, 2008.
[19] Yogesh S, Pradeep K, and O. S, "A Review of Studies On Machine Learning Techniques," International
Journal of Computer Science and Security vol. 1, pp. 70-84.
[20] Nguyen V, "The Application of Machine Learning Methods in Software Verification and Validation " MSc,
Master of Science in Engineering, The University of Texas at Austin, Texas 2010.
[21] Ekbal R, Srikanta P, and V. B, "Software Quality Estimation using Machine Learning:Case-based Reasoning
Technique " International Journal of Computer Applications, vol. 58, pp. 43-48, November 2012.
[22] E. Erturk and E. A. Sezer, "A comparison of some soft computing methods for software fault prediction,"
Expert Systems with Applications, 2014.
[23] M. SURENDRA and D. N. GEETHANJALI, "Classification of Defects In Software Using Decision Tree
Algorithm," vol. 5, pp. 1332-1340, June 2013.
[24] Saiqa A, Luiz F, and F. A, "Benchmarking Machine Learning Techniques for Software Defect Detection,"
International Journal of Software Engineering & Applications, vol. 6, pp. 11-23, May 2015.
[25] Martin S, David B, and T. H, "Researcher Bias: The Use of Machine Learning in Software Defect Prediction "
IEEE TRANSACTIONS ON SOFTWARE ENGINEERING, vol. 40, pp. 603-616, JUNE 2014.
[26] X. Tan, X. Peng, S. Pan, and W. Zhao, "Assessing Software Quality by Program Clustering and Defect
Prediction," pp. 244-248, 2011.
[27] Jaspree K and P. S, "A k-means Based Approach for Prediction of Level of Severity of Faults in Software
System," Proceedings of International conference on Intelligent Computational Systems, 2011.
[28] Pooja P and D. A. Phalke, "Software Defect Prediction for Quality Improvement Using Hybrid Approach,"
International Journal of Application or Innovation in Engineering & Management, vol. 4, June 2015.
[29] A. Kaur and R. Malhotra, "Application of Random Forest in Predicting Fault-Prone Classes," pp. 37-43, 2008.
[30] Y. I. Peng, G. Kou, G. Wang, W. Wu, and Y. Shi, "Ensemble of Software Defect Predictors: An Ahp-Based
Evaluation Method," International Journal of Information Technology & Decision Making, vol. 10, pp. 187-
206, 2011.

© www.ermt.net All Rights Reserved Page | 306

View publication stats