An empirical evaluation of the effectiveness of inspection scenarios developed from a defect repository

Abstracting and summarizing high-severity defects detected during inspections of previous software versions could lead to effective inspection scenarios in a subsequent version in software maintenance and evolution. We conducted an empirical evaluation of 456 defects detected from the requirement specification inspections conducted during the development of industrial software. The defects were collected from an earlier version, which included 59 high-severity defects, and from a later version, which included 48 high-severity defects. The results of the evaluation showed that nine defect types and their corresponding inspection scenarios were obtained by abstracting and summarizing 45 defects in the earlier version. The results of the evaluation also showed that 46 of the high-severity defects in the later version could be potentially detected using the obtained inspection scenarios. The study also investigated which inspection scenarios can be obtained by the checklist proposed in the value-based review (VBR). It was difficult to obtain five of the inspection scenarios using the VBR checklist. Furthermore, to investigate the effectiveness of cluster analysis for inspection scenario development, the 59 high-severity defects in the earlier version were clustered into similar defect groups by a clustering algorithm. The results indicated that cluster analysis can be a guide for selecting similar defects and help in the tasks of abstracting and summarizing defects.