A Comprehensive Study on Ensemble Methods for Software Error Detection

Abstract

Software Developers often find it painful and tedious to locate or pinpoint the errors that reside in the source code which causes serious hindrance in the progression of developing any software. In the field of software engineering, it is very crucial to understand the software metrics that are directly involved with the progression of the software. Besides, various classification algorithms have been used to foresee the errors in building the software. In this paper, we focus especially on ensemble algorithms as they tend to provide more precise and statistically efficient outcomes than the other traditional algorithms. This paper contains twenty software metrics that are pivotal in identifying errors in software applications. Eight Java projects have been gathered to showcase the significance of the software metrics in predicting errors. In this study, three ensemble methods are considered, MultiBoostAB, Dagging, and Decorate. For a detailed inspection of the performance, accuracy, recall, precision, F-measure, and ROC Curve were appraised. The comparisons exhibit Decorate as the highest-performing method and Dagging as the lowest.