Empirical Analysis of Ensemble Machine Learning Techniques for Bug Triaging

Abstract

Bug reports are an inescapable part of the software product framework. Nowadays, software advancements have led to the creation of beta versions of software in order to assemble the bug reports from clients. The assembled bug reports are then handled by software developers to make consequent software more reliable as well as robust. However, high recurrence of approaching bug reports forges the process of bug fixing to be a troublesome & tedious process. Bug triaging is an essential component of issue handling process and it deals with the selection of a suitable software developer for handling of reported bug such that the assigned developer is able to fix the reported issue. In the literature, different semi and fully mechanized procedures are proposed to facilitate the endeavor of developer selection in bug repositories. These techniques use historically fixed information from bug repositories to classify any new incoming bugs. In the recent years, ensemble-based classification techniques have gained popularity. These techniques use multiple classifiers for making a prediction and has proved to be outperforming classical machine learning classification. In this paper, we present an empirical study of ensemble-based techniques for classification of new incoming bug reports. We studied 5 ensemble classification techniques, namely Bagging, Boosting, Majority Voting, Average Voting, and Stacking using 25 different machine learning classifiers as base classifiers. The experimental results showed that ensemble classifiers outperform classical machine learning algorithms for selection of suitable developer for handling the bug report.