Research on Cross-version Software Defect Prediction Based on Evolutionary Information

Abstract

Software defect prediction is an important part of the software testing field. According to the characteristics of object-oriented software, this paper considers the evolution information separately in different packages and summarizes the evolution metrics that affect the defect prediction. Existing research on evolutionary information often ignores the impact of newly added and disappearing classes on software defects prediction. Based on these factors, evolution metrics are proposed and applied to defect prediction. Two evolution metrics, transition class ratio and static metric category number, are proposed for object-oriented cross-version defect prediction. Experiments are carried out in the commonly used software defect prediction set. The experimental results show that the proposed metrics have better defect prediction ability than the traditional static metric.