Prediction of Pull Requests Review Time in Open Source Projects

In open-source projects that receive large amounts of pull requests, the tasks of maintaining quality and prioritizing code review have become a complex task. In this sense, several works explored data on pull requests in order to provide useful information. Although, the review time was treated as the interval between the submission and the integration of the pull requests, that is, the lifetime itself. Since December 2016, a feature allows requesting reviews to one or more specific reviewers, which, together with the review status, allowed to establish the period closest to the effective code review time, the interval between the review request and the last review with approval status. In this context, the main objective of this work is to predict the review time of pull requests. Furthermore, the lifetime and acceptance of pull requests with and without review time were compared and the CFS (Correlation-based Feature Selection) attribute selection strategy was used to identify those most relevant to the forecast. The results of the experiments indicate that the SMO (Sequential Minimal Optimization) algorithm had the smallest error, averaging 8,504 minutes (approximately 5,9 days) and that the presence of approvals in the review requests has a positive influence on both the acceptance and in the pull request lifetime.