Comparative Analysis of Software Network and CK Metrics: Implications for Pre- and Postrelease Faults

摘要

The research on open source software hasattracted a great deal of attention during the past decadesfor its wide applications in both academia and industry.Among the research topics related to open source software,the usefulness of software network metrics for faultprediction has been received much attention recently. Inorder to verify the importance of software network metricsin the performance of fault prediction models, this studymixes software network and Chidamber & Kemerer (CK)metrics to predict pre- and post-release fault-proneness ofEclipse version 2.0. The "small world" and "scale-free"characteristics and Spearman correlation have beenanalyzed in detail. In addition, we employed10 classifiers, 2weak classifiers, and 5 performance evaluation criteria toconduct three sets of experiments. Experimental resultshave demonstrated that the software network metrics canonly provide little overall performance improvement. Basedon the comprehensive study, the effect of software networkmetrics can be ignored.