Blocking Bug Prediction Based on XGBoost with Enhanced Features

With a growing number of software projects, software quality is increasingly crucial. Researchers and engineers in the software engineering field often pay much attention to bug management tasks, such as bug localization, bug triage, and duplicate bug detection. However, there are few researchers to study blocking bug prediction. Blocking bugs prevent other bugs from being fixed and usually need more time to be fixed. Thus, developers need to identify blocking bugs and reduce the impact of blocking bugs. The previous studies utilized supervised algorithms to implement this task. However, they did not consider the dependencies among individual classifiers so that they cannot get the perfect accuracy for blocking bug prediction. In this paper, we propose a new framework XGBlocker that includes two stages. In the first stage, XGBlocker collects more features from bug reports to build an enhanced dataset. In the second stage, XGBlocker exploits XGBoost technique to construct an effective model to perform the prediction task. We conduct experiments on four projects with three evaluation metrics. The experimental results show that our method XGBlocker achieves promising performance compared with baseline methods in most cases. In detail, XGBlocker achieves F1-score, ER@20%, and AUC of up to 0.808, 0.944, and 0.975, respectively. On average across the four projects, XGBlocker improves F1-score, ER@20%, and AUC over the state-of-the-art method ELBlocker by 17.27%, 12.67%, and 4.85%, respectively.