Blocking Bugs Identification via Binary Relevance and Logistic Regression Analysis

Blocking bugs, a type of bugs that prevents other bugs from being fixed, significantly increase the fixed time of both themself and the blocked bugs. Thus, these blocking bugs bring a considerable negative impact on software evolution. Therefore, the timely identification of blocking bugs is essential for software maintenance. This paper proposes an approach based on Binary Relevance(BR) and Logistic Regression(LR) analysis, called BR-LR, to predict bugs' blocking and blocked labels. We first filter and build a dataset consisting of two sets with a specific type of blocking relationship based on the ideas of BR. Then, we extract several fields from the bug reports and train the model by applying the logistic regression analysis with the constructed dataset in the first step, resulting in two prediction models for bug blocked and blocking labels. Finally, our approach combines the two prediction results to identify whether the bug is blocking or blocked. We also conduct empirical studies on seven open-source projects to verify the effectiveness of our approach. The final experimental results show that our model performs better from a partially correct perspective and can accurately predict bug labels than benchmarks. Specifically, the average accuracy of our model is 54.86%, and the average F1-measure is 50.61 %.