Predicting priority level of bug reports

• Main Author: Ang, Zi Qun
• Other Authors: Xing Zhenchang
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/66813
• Institution: Nanyang Technological University

A bug repository stores numerous bugs reported for a software development projects. The bug repository is usually maintained in the software development projects. The software developers would access the bug repository and investigate the reported bugs, so as to improve the software quality. Bug reports are reported by the users when bugs are found while using the software, and submitted to bug tracking system like Bugzilla. A bug report submitted by a user would contain the following information; the summary and the description of the bug, the product and the component affected by the bug, and the severity level of the bug. Bugs being inevitable and are found frequently, bugs reported are required to be priorities. A bug triager’s task would be to assign a priority level to the reported bugs based on the submitted information. The software developers would then investigate the reported bugs based on their importance, determined by the assigned priority level. However, assigning a priority level to a bug report is done manually. Considering the fact that in practice, bug reports are reported daily, it can be time consuming for the bug triager to manually assign a priority level to the reported bugs. Thus, to address this issue, this project aims to develop an automated approach that would suggest a priority level, depending on the submitted information in a bug report. The approach uses machine learning to achieve the aim, and by considering the following features; summary, description, product, component and severity level that could possibly affects the priority level assigned to a bug report. These features are extracted, and fed into three different classifiers, namely; Naïve Bayes Multinomial, Stochastic Gradient Descent, and Support Vector Machine, which are then used to train a discriminative model. The experiments are evaluated on bug reports from the Eclipse datasets, a large open source projects from bug tracking system Bugzilla. The results shows that Support Vector Machine performs the best compared to the other two classifiers with accuracy score of 51.20%, outperforming Naïve Bayes Multinomial classifier and Stochastic Gradient Descent classifier by 18.40% and 17.10% respectively.