People Profile Metrics for Improved Classification of Defect Prone Files in Open Source Projects

Numerous models have been studied and presented in literature for classification of defect-prone source code files. Usually these models use static code metrics, process metrics, and change metrics as input and predict defect proneness of code. However, there has been limited use of people related metrics as input to the prediction models. Impact of using people related information should be studied for better classification of defect prone files in future releases of software projects. This study proposes the use of People Profile Metrics (PPM) of software development team members to improve the prediction of defect prone source code files. The experiment is performed on an open source project and the defect prone source code files have been classified. In addition, severity of defects has also been predicted. The PPM have been evaluated for three classifiers Decision Tree, Random Forest, and K-Nearest Neighbors using Weka. Significant improvement in classification of defect prone source code files, in terms of Precision, Recall and F-Measure has been achieved. The combination of existing static code metrics and the PPM will be tested for more projects and for unsupervised models.