Convolution neural network based priority prediction approach for GITHUB issues

Abstract

A bug report is an important document that outlines software problems that result in unexpected errors or wrong outcomes. In large software projects, a high number of bugs are reported daily, which needs to be systematically analyzed. Predicting the priority level of reported bugs, assigning an appropriate developer, finding duplicate issues, and predicting bug resolving time are some of the critical tasks in the bug analysis process. Due to the inherent complexity of the bug analyzing process, manual bug investigation requires a significant amount of time, resources, and effort. Therefore, the need to establish automated or semi-automated approaches for assessing bug reports is extensively discussed in the literature. This research presents a novel approach to prioritize the bug reports by exploiting a Convolutional Neural Network-based approach. Furthermore, this research investigates the impact of both textual and categorical features of bug reports in improving the accuracy of priority prediction. The experiments were conducted by extracting the bug reports available in three GitHub repositories. The evaluation results confirm that the use of categorical features does not have an impact on the accuracy of the priority prediction of bug reports. Furthermore, it was observed that better prediction accuracies are shown for the datasets extracted from Bugzilla than GitHub repository