Evaluating Performance of Network Metrics for Bug Prediction in Software

Abstract

Code-based metrics and network analysis based metrics are widely used to predict defects in software. However, their effectiveness in predicting bugs either individually or together is still actively researched. In this paper, we evaluate the performance of these metrics using three different techniques, namely, Logistic regression, Support vector machines and Random forests. We analysed the performance of these techniques under three different scenarios on a large dataset. The results show that code metrics outperform network metrics and also no considerable advantage in using both of them together. Further, an analysis on the influence of individual metrics for prediction of bugs shows that network metrics (except out-degree) are uninfluential.