Automated labelling and severity prediction of software bug reports

摘要

Our main aim is to develop an intelligent classifier that is capable of predicting the severity and label (type) of a newly submitted bug report through a bug tracking system. For this purpose, we build two datasets that are based on 350 bug reports from the open-source community (Eclipse, Mozilla, and Gnome). These datasets are characterised with various textual features. Based on this information, we train variety of discriminative models that are used for automated labelling and severity prediction of a newly submitted bug report. A boosting algorithm is also implemented for an enhanced performance. The classification performance is measured using accuracy and a set of other measures. For automated labelling, the accuracy reaches around 91% with the AdaBoost algorithm and cross validation test. On the other hand, for severity prediction, the classification accuracy reaches around 67% with the AdaBoost algorithm and cross validation test. Overall, the results are encouraging.