Defects in The Next Release; Software Defect Prediction Based on Source Code Versions

Nowadays, automatic defect prediction has attracted wide attention, but many researchers neglected the availability of huge amount of data in version control systems. Deploying the next release of the project on time and on schedule urges software project managers to optimize efforts based on their limited human resources. Usually software projects have longterm version history from which we can build a repository of rnetrics, defects, fixes and causes. Having an accurate estimate of the distribution of bugs across components helps project managers to optimize the available resources by focusing on the problematic system parts. We have introduced a benchmark for interested readers in this field which provides all the data needed to apply a large array of prediction techniques proposed in the literature. Our dataset allows the reproduction of the experiments reported in this paper and their comparison with novel defect prediction approaches. We evaluated a selection of representative classification approaches and compared the performance of major classifiers using various evaluation measures such as Precision, Recall, F-measure, G-mean and AUC. The experimental results show that Random Forest is a powerful method and as a good candidate for software defect prediction can be used for further improvements.