Authors:
Lianfa Li
1
and
Hareton Leung
2
Affiliations:
1
CAS and The Hong Kong Polytechnic University, China
;
2
The Hong Kong Polytechnic University, China
Keyword(s):
Object-Oriented Systems, Fault-proneness, Software Quality, Data Mining.
Related
Ontology
Subjects/Areas/Topics:
Enterprise Information Systems
;
Information Systems Analysis and Specification
;
Software Engineering
;
Software Metrics and Measurement
Abstract:
In the prediction of fault-proneness in object-oriented (OO) systems, it is essential to have a good prediction method and a set of informative predictive factors. Although logistic regression (LR) and naïve Bayes (NB) have been used successfully for prediction of fault-proneness, they have some shortcomings. In this paper, we proposed the Bayesian network (BN) with data mining techniques as a predictive model. Based on the Chidamber and Kemerer’s (C-K) metric suite and the cyclomatic complexity metrics, we examine the difference in the performance of LR, NB and BN models for the fault-proneness prediction at the class level in continual releases (five versions) of Rhino, an open-source implementation of JavaScript written in Java. From the viewpoint of modern software development, Rhino uses a highly iterative or agile development methodology. Our study demonstrates that the proposed BN can achieve a better prediction than LR and NB for the agile software.