CIL-BSP: Bug Report Severity Prediction based on Class Imbalanced Learning

Abstract

The bug reports' severity can be used for developers to prioritize which bugs to be fixed first. However, this process depends on the developers' expertise in assigning the correct bug severity. In our previous study, we propose a novel bug report severity prediction method EKD-BSP, which utilizes the bug summary and the keywords extracted from the bug description. However, a class imbalance exists in our gathered bug report severity prediction datasets. To solve this issue, we design a novel method CIL-BSP by further considering the class imbalanced methods. Moreover, we apply hyperparameter optimization to CIL-BSP and consider different optimization strategies. To verify the effectiveness of CIL-BSP, we select two real-world open-source projects Eclipse and Mozilla as our experimental subjects. Based on our empirical results, we find that performing hyper-parameter optimization can significantly improve the severity prediction performance of CIL-BSP. Moreover, optimization hyperparameters on the classifier can contribute more than the class imbalanced method.