AI-Assisted Bug Prediction and Code Review Automation using Machine Learning

This paper proposes an AI-aided framework for bug-prone commit prediction and automated code review, leveraging a hybrid of traditional machine learning with deep learning models. We harness commit-level datasets such as JIT, DeepJIT, and state-of-the-art high-confidence benchmarks, integrate process and semantic features, and evaluate models like Random Forest, XGBoost, LSTM, and CodeBERT. Explainability is ensured through SHAP, while model calibration techniques have also been applied to make the probability outputs trustworthy for any real-world application. We report experimental results, discuss deployment in CI/CD, and summarize recent advances through 20 recent research studies in the area from 2023 to 2025. Moreover, the proposed framework underscores scalability and adaptability with any type of programming language and repository. The system provides actionable insights to developers through visual dashboards and automated alerts, enhancing decision-making during code reviews. Finally, the study underlines potential extensions toward reinforcement learning and human-in-the-loop collaboration for continuous improvement of the models themselves and trust enhancement