Impact of Types of Change on Software Defect Prediction

Churn metrics are commonly used in software defect prediction due to high performance, language independence and ease of extraction. The main data sources for churn metrics-based prediction models in the literature are the change logs. But changes may also be due to several reasons other than fixing defects, such as enhancement and applying new requirements. Therefore, without awareness of type of the change and effects on defect prediction, fitting the best model is too difficult. In this paper, we propose a churn type-aware defect prediction model. We observe the impacts of churn-type awareness on prediction accuracy level. In our experiments, we used a real work item and change log dataset which is produced by a software development team. In our study, we also reveal the correlations among the change types.