Learning to Predict Code Review Rounds in Modern Code Review Using Multi-Objective Genetic Programming
Moataz Chouchen, Issam Oukhay, Ali Ouni · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2025
Code review is an essential practice for software quality assurance. However, code review can be cumbersome as patches often undergo multiple rounds to fix bugs, enforce coding standards, and improve structure before merging or abandonment. Predicting the number of review rounds can help developers prioritize tasks and streamline the process. Existing machine learning models for review round prediction suffer from key limitations. Their black-box nature makes them difficult to interpret, reducing trust and adoption. Additionally, they rely on data re-balancing techniques that introduce artificial points, causing concept shifts and reducing reliability. To address these issues, we propose MORRP, a novel Multi-Objective Review Rounds Prediction approach. MORRP is based on Multi-Objective Genetic Programming (MOGP) to predict review rounds. Our method evolves interpretable models while optimizing precision, recall, and specificity without relying on data re-balancing. We evaluate our approach on three large open-source projects: Eclipse, OpenDaylight, and OpenStack. Results show that MORRP achieves competitive performance, with a micro F1 score between 0.65 and 0.75, outperforming complex ML models like Random Forest and LightGBM.