Art of software defect association & correction using association rule mining

International Journal of Computer and Technology (IJCET), ISSN 0976 – 6367(Print),
International Journal of Computer Engineering Engineering
ISSN 0976 – 6375(Online) Volume 1, Number 1, May - June (2010), © IAEME
and Technology (IJCET), ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online) Volume 1 IJCET
Number 1, May - June (2010), pp. 261-271 ©IAEME
© IAEME, http://www.iaeme.com/ijcet.html

ART OF SOFTWARE DEFECT ASSOCIATION &
CORRECTION USING ASSOCIATION RULE MINING
Nilamadhab Mishra, MCA, M.PHIL (C S), M.Tech (CSE)
Asst.prof, Krupajal Group of Institution, Orissa
E-mail : nilamadhab.mishra@rediffmail.com

ABSTRACT:
Much current software defect prediction work focuses on the number of defects
remaining in a software system. In this paper, I focus about association rule mining based
methods to predict defect associations and defect correction effort. This is to help
developers detect software defects and assist project managers in allocating testing
resources more effectively. Software process improvement method is based on the
analysis of defect data. I applied the proposed methods to the SEL defect data consisting
of more than 200 projects over more than 15 years. The method is classification of
software defects using attributes which relate defects to specific process activities. The
results show that, for defect association prediction, the accuracy is very high and the
false-negative rate is very low. The comparison of defect correction effort prediction
method with other types of methods—PART, C4.5, and Naı¨ve Bays show that accuracy
can be improved. I also evaluate the impact of support and confidence levels on
prediction accuracy, false negative rate, false-positive rate, and the number of rules. Thus
higher support and confidence levels may not result in higher prediction accuracy and a
sufficient number of rules must be a precondition for high prediction accuracy.
Key words: Defect association, defect correction effort, defect isolation effort, defect
association prediction, Software defect prediction
1. INTRODUCTION
The success of a software system depends not only on cost and schedule, but also
on quality. Among many software quality characteristics, residual defects have become
the de facto industry standard [1, 2]. Therefore, the prediction of software defects, i.e.,

261

International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 – 6367(Print),

deviations from specifications or expectations which might lead to failures in operation,
has been an important research topic in the field of software engineering for more than 30
years. Clearly, they are a proxy for reliability, but, unfortunately, reliability is extremely
difficult to assess prior to full deployment. Current defect prediction work focuses on
estimating the number of defects remaining in software systems with code metrics,
inspection data, and process-quality data by statistical approaches, capture-recapture
(CR) models, and detection profile methods (DPM)[4 ].
The prediction result, which is the number of defects remaining in a software
system, can be used as an important measure for the software developer, and can be used
to control the software process and gauge the likely delivered quality of a software
system. In contrast, propose that the defects found during production are a manifestation
of process deficiencies, so they present a case study of the use of a defect based method
for software in-process improvement. In particular, they use an attribute-focusing (AF)
method to discover associations among defect attributes such as defect type, source, and
phase introduced, phase found, component, impact, etc [9]. By finding out the event that
could have led to the associations, they identify a process problem and implement a
corrective action. This can lead a project team to improve its process during
development. I restrict my work to the predictions of defect (type) associations and
corresponding defect correction effort.
1. For the given defect(s), what other defect(s) may co-occur?
2. In order to correct the defect(s), how much effort will be consumed?
I use defect type data to predict software defect associations that are the relations
among different defect types such as: If defects a and b occur, then defect c also will
occur. This is formally written as a ^ b => c. The defect Associations can be used for
three purposes [3, 4].
First, find as many related defects as possible to the detected defect(s) and,
consequently, make more-effective corrections to the software. For example, consider the
situation where we have classes of defect a, b, and c and suppose the rule a ^ b => c has
been obtained from a historical data set, and the defects of class a and b have been
detected occurring together, but no defect of class c has yet been discovered. The rule
indicates that a defect of class c is likely to have occurred as well and indicates that we

262


should check the corresponding software artifact to see whether or not such a defect
really exists. If the result is positive, the search can be continued, if rule a ^ b ^ c =>d
holds as well, we can do the same thing for defect d.
Second, help evaluate reviewers’ results during an inspection. For example, if rule
a ^ b => c holds but a reviewer has only found defects a and b, it is possible that he
missed defect c. Thus, a recommendation might be that his/her work should be reinserted
for completeness.
Third, to assist managers in improving the software process through analysis of
the reasons some defects Frequently occur together. If the analysis leads to the
identification of a process problem, managers have to come up with a corrective action.
At the same time, for each of the associated defects, we also predict the likely
effort required to isolate and correct it. This can be used to help project managers
improve control of project schedules. Both of our defect association prediction and defect
correction effort prediction methods are based on the association rule mining method
which was first explored by Agrawal. Association rule mining aims to discover the
patterns of co-occurrences of the attributes in a database [5, 6]. However, it must be
stressed that associations do not imply causality. An association rule is an expression A--
>C, where A (Antecedent) and C (Consequent) are sets of items. The meaning of such
rules is quite intuitive: Given a database D of transactions, where each transaction T 2D
is a set of items, A-->C expresses that whenever a transaction T contains A, then T also
contains C with a specified confidence. The rule confidence is defined as the percentage
of transactions containing C in addition to A with regard to the overall number of
transactions containing A. The idea of mining association rules originates from the
analysis of market-basket data where rules like “customers who buy products p1 and p2
will also buy product p3 with probability c percent” are extracted. Their direct
applicability to business problems together with their inherent understandability make
association rule mining a popular data mining method.

263


2. PAPER CONTENTS
General method:
The objective of the study is to discover software defect associations from
historical software engineering data sets, and help determine whether or not a defect(s) is
accompanied by other defect(s). If so, we attempt to determine what these defects are and
how much effort might be expected to be used when we correct them. Finally, we aim to
help detect software defects and effectively improve software control. For this purpose,
first, we preprocessed the NASA SEL data set (see the following section for details), and
obtained three data sets: the defect data set, the defect isolation effort data set, and the
defect correction effort data set. Then, for each of these data sets, we randomly extracted
five pairs of training and test data sets as the basis of the research. After that, we used
association rule mining based methods to discover defect associations and the patterns
between a defect and the corresponding defect isolation/correction effort. Finally, we
predicted the defect(s) attached to the given defect(s) and the effort used to correct each
of these defects. We also compared the results with alternative methods where applicable.
Data Source and Data Extraction:
The data we used is SEL Data which is a subset of the online database created by
the NASA/GSFC Software Engineering Laboratory (SEL) for storage and retrieval of
software engineering data for NASA Goddard Space Flight Center. The subset includes
defect data of more than 200 projects completed over more than 15 years. The SEL Data
is a database consisting of 15 tables that provide data on the projects’ software
characteristics (overall and at a component level), changes and errors during all phases of
development, and effort and computer resources used. In the SEL Data, the defects are
divided into six types, Table 1 contains the details. In addition, the effort used to correct
defects falls into four categories: One Hour or Less, One Hour to One Day, One Day to
Three Days, and More Than Three Days.
Analysis approach:
I use the five-fold cross-validation method as the overall analysis approach. That
is, for each D of the defect data set, the defect isolation effort data set, and the defect
correction effort data set, the inducer is trained and tested a total of five times. I use the

264


association rule mining method to learn rules from the training data sets. For defect
association prediction, the rule learning is straightforward, while for defect correction
effort prediction, it is more complicated because the consequent of a rule has to be defect
correction effort [7, 8]. Considering the target of association rule mining is not
predetermined and classification rule mining has only one predetermined target, the class,
I integrate these two techniques to learn defect correction effort prediction rules by
focusing on a special subset of association rules whose consequents are restricted to the
special attribute, the effort. Once we obtain the rules, we rank them, and use them to
predict the defect associations and defect correction effort with the corresponding test
data sets. The predictions of defect associations and defect correction effort are both
based on the length-first (in terms of the number of items in a rule) strategy.
Association Rule Discovery:
Both positive and negative association rules are to be discovered by the project.
First, the frequent item sets of potential interest and infrequent item sets of potential
interest are generated in a database. Then Extract positive rules of the form A -->B, and
negative rules of the forms A-->¬B, ¬A-->B, and ¬A ¬B.
Rule ranking Strategy:
Before prediction, we rank the discovered rules according to the length-first
strategy. The length-first strategy was used for two reasons. First, for the defect
association prediction, the length-first strategy enables us to find out as many defects as
possible that coincide with known defect(s), thus preventing errors due to incomplete
discovery of defect associations. Second, for the defect correction effort prediction, the
length-first strategy enables us to obtain more-accurate rules, thus improving the effort
prediction accuracy. Specifically, the length-first rule-ranking strategy is as follows:
1. Rank rules according to their length. The longer the rules, the higher the priority.
2. If two rules have the same length, rank them according to their confidence values. The
greater the confidence values, the higher the priority. The more-confident rules have
more predictive power in terms of accuracy; thus, they should have higher priority.

265


3. If two rules have the same confidence values, rank them according to their support
values. The higher the support values, the higher the priority. The rules with higher
support value are more statistically Significant, so they should have higher priority.
4. If two rules have the same support value, rank them in alphabetical order.
2.1 GRAPHS, TABLES, AND PHOTOGRAPHS
Defect Types:
Here we are concentrated on some specified defects that are the most frequent
defects that may occur in the Sel data.
Defect Types:

266


Rule Ranking Algorithm:

Defect Isolation and Correction Effort Prediction:
The constraint-based association rules mining method is used for the purpose of defect
isolation and correction prediction [10]. The steps are
1. Compute the frequent item sets that occur together in the training data set at least as
frequently as a predetermined min: Supp. The item sets mined must also contain the
effort labels
2. Generate association rules from the frequent item sets, where the consequent of the
rules is the effort. In addition to the min: Supp threshold, these rules must also satisfy a
minimum confidence.

267


Effort Prediction Rules:

Defect association prediction procedure.
We use the association rule mining method to learn rules from the training data
sets. For defect association prediction, the rule learning is straightforward, while for
defect correction effort prediction, it is more complicated because the consequent of a

268


rule has to be defect correction effort. Considering the target of association rule mining is
not predetermined and classification rule mining has only one predetermined target, the
class, we integrate these two techniques to learn defect correction effort prediction rules
by focusing on a special subset of association rules whose consequents are restricted to
the special attribute, the effort. Once we obtain the rules, we rank them, and use them to
predict the defect associations and defect correction effort with the corresponding test
data sets. The predictions of defect associations and defect correction effort are both
based on the length-first strategy.
There is no work on software defect association prediction and there is no other
method that can be used for this purpose. Therefore, I am unable to directly compare our
defect association prediction method with other studies. As the defect correction effort is
represented in the form of categorical values and there are some attributes to characterize
it, it can be viewed as a classification problem. This allows us to compare our defect
correction prediction method with three different types of methods. These methods are
the Bayesian rule of conditional probability-based method, Naı¨ve Bayes [, the well-
known trees-based method, C4.5, and the simple and effective rules-based method,
PART.
3. CONCLUSION
In this paper, I have given an application of association rule mining to predict
software defect associations and defect correction effort with SEL defect data. This is
important in order to help developers detect software defects and project managers
improve software control and allocate their testing resources effectively. The ideas have
been tested using the NASA SEL defect data set. From this, I extracted defect data and
the corresponding defect isolation and correction effort data.
I have also compared the defect correction effort prediction method with three
other types of machine learning methods, namely, PART, C4.5, and Naı¨ve Bayes. The
results show that for defect isolation effort, our accuracy is higher than for the other three
methods b. Likewise, for defect correction effort prediction, the accuracy is higher than
the other three methods.

269


I also have explored the impact of support and confidence levels on prediction
accuracy, false negative rate, false positive rate, and the number of rules as well. I found
that higher support and confidence levels may not result in higher prediction accuracy,
and a sufficient number of rules is a precondition for high prediction accuracy. While we
do not wish to draw strong conclusions from a single data set study, i believe that my
results suggest that association rule mining may be an attractive technique to the software
engineering community due to its relative simplicity, transparency, and seeming
effectiveness in constructing prediction systems.
REFERENCES:
[1]. Inderpal S. Bhandari, M. J. Halliday, E. Tarver, D. Brown, J. Chaar, and R.
Chillarence. A Case Study of Software Process Improvement during Development. IEEE
Trans. on Soft. Eng., 19(12), pp. 1157-1170, December 1993.
[2] Inderpal S. Bhandari, M. J. Halliday, J. Chaar, R. Chillarence, K. Jones, J. S.
Atkinson, C. Lepori-Costello, P. Y. Jasper, E. D. Tarver, C. C. Lewis, and M. Yonezawa.
In-Process Improvement through Defect Data Interpretation. IBM Syst. Journal, 33(1),
pp. 182-214, 1994.
[3] Xindong Wu, Chengqi Zhang, Shichao Zhang,” Efficient Mining of both positive and
negative association rules”, ACM Transactions on Information Systems, Vol. 22, No. 3,
July 2004, Pages 381–405.
[4] Qinbao Song, Martin Shepperd, Michelle Cartwright, Carolyn Mair, "Software Defect
Association Mining and Defect Correction Effort Prediction," IEEE Transactions on
Software Engineering, vol. 32, no. 2, pp. 69-82, Feb., 2006.
[5] I.S. Bhandari, M.J. Halliday, J. Chaar, R. Chillarenge, K. Jones, J.S. Atkinson, C.
Lepori-Costello, P.Y. Jasper, E.D. Tarver, C.C. Lewis, and M. Yonezawa, “In Process
Improvement through Defect Data Interpretation,” IBM System J., vol. 33, no. 1, pp.
182-214, 1994.
[6] L.C. Briand, K. El-Emam, B.G. Freimut, and O. Laitenberger, “A Comprehensive
Evaluation of Capture-Recapture Models forEstimating Software Defect Content,” IEEE
Trans. Software Eng., vol. 26, no. 6, pp. 518-540, June 2000.

270


[7] T. Compton and C. Withrow, “Prediction and Control of Ada Software Defects,” J.
Systems and Software, vol. 12, pp. 199-207, 1990.
[8] F. Brito e Abreu, Melo, W., "Evaluating the Impact of Object-Oriented Design on
Software Quality", Proceedings of Third International Software Metrics Symposium, pp.
90- 99, 1996.
[9] S. R. Chidamber and C. F. Kemerer, "A Metrics Suite for Object Oriented Design",
IEEE Transactions on Software Engineering, 20(6), pp. 476-493, 1994.
[10] D. ubrani , Murphy, G.C., "Hipikat: recommending pertinent software development
artifacts", Proceedings of International Conference on Software Engineering, pp. 408-
418, 2003.

271

Art of software defect association & correction using association rule mining

More Related Content

What's hot (15)

Viewers also liked (20)

Similar to Art of software defect association & correction using association rule mining (20)

More from IAEME Publication (20)

Art of software defect association & correction using association rule mining