Cross-Version Defect Prediction using Cross-Project Defect Prediction Approaches

Abstract

Background: Specifying and removing defects before release deserve extra cost for the success of software projects. Long-running projects experience multiple releases, and it is a natural choice to adopt cross-version defect prediction (CVDP) that uses information from older versions. A past study shows that feeding multi older versions data may have a positive influence on the performance. The study also suggests that cross-project defect prediction (CPDP) may fit the situation but one CPDP approach was only examined.Aims: To investigate whether feeding multiple older versions data is effective for CVDP using CPDP approaches. The investigation also involves performance comparisons of the CPDP approaches under CVDP situation. Method: We chose a style of replication of the comparative study on CPDP approaches by Herbold et al. under CVDP situation. Results: Feeding multiple older versions had a positive effect for more than a half CPDP approaches. However, almost all of the CPDP approaches did not perform significantly better than a simple rule-based prediction. Although the best CPDP approach could work better than it and with-in project defect prediction, we found no effect of feeding multiple older versions for it. Conclusions: Feeding multiple older versions could improve CPDP approaches under CVDP situation. However, it did not work for the best CPDP approach in the study.