Utilizing feature based classification and textual information of bug reports for severity prediction

Predicting bug severity is important task in software development and maintenance. If bug severity is predicted accurately, it would be a significant assistance for software developers to allocate resource and fix bugs. Thus, this paper presents a way to predict bug severity. First, this study trains bugs in bug repository which stores previously reported bug reports. This study applies feature based classification using meta-fields in bug reports in training where Multinomial Naive Bayes(MNB) is used. Next, when new bug is reported, this study predicts its severity by using textual information (Summary, Description) of new bug and existing bugs. We evaluate the performance of our method using two large-scale open-source projects, including Eclipse, and Mozilla. The experimental results reveal that our approach outperforms other severity prediction method.