A Fault-prone Module Prediction Combining Past and Ongoing Test Results

Abstract

The aim of this study is to improve the efficiency of software testing and/or review by predicting modules that are likely to contain a bug. In cross-version bug prediction, which uses bug data from a previous version to predict bugs in software currently under development, it is difficult to achieve high prediction accuracy due to differences in bug factors between versions. In this research, three methods (test method, mix method, select method) are proposed to address this challenge by incorporating ongoing test results, in addition to bug data from past versions. An evaluation experiment targeted 13 open-source software projects, comparing conventional cross-version prediction with the three proposed methods using seven evaluation metrics. The result indicates that each proposed method achieves higher accuracy in bug prediction than the conventional cross-version prediction when utilizing test results for approximately 10% of the modules in the developing software. Particularly, the select method demonstrated superior performance in comparison.