(An) effective software defect prediction based on change level defect removal activity

Abstract

Software defect prediction is a technique for statistically identifying defective modules or files. Recently, as the open source and version control systems are popular, defect prediction techniques of change units are actively studied. However, the metrics used in change level defect prediction only considered defect induction aspects and did not consider defect removal activity. Defect removal activity is the activity that eliminates defects through automated analysis, testing and peer review. Defect removal activity is one of the important software quality assurance activities that reduce the residual defects of software. In this paper, we propose metrics for quantifying defect removal activity in change level environment and evaluate the performance improvement of the defect prediction model using the metrics. The defect prediction model with the new metrics showed better performance than the defect prediction model using only the existing metric.