Detail

Can Commit History Predict Future Code Changes in GitHub Projects?

During software maintenance and evolution, a challenging problem for development teams is the management and resourcing of future software changes. These software changes can range from planned feature enhancements and optimizations to unplanned patches of critical bugs. In our research we explore the role automation can play in assisting with the management of future code changes. Our approach uses machine learning (support vector machines) trained with historical GitHub commit data to predict the sections of source code that are likely to be changed in the near future. In order to understand the feasibility of our approach we conducted experiments on 23 mature GitHub projects. Furthermore, our experiments explore the impact that the feature set, the sampling range, and the use of balanced sampling have on the accuracy, precision and recall of our predictions.