Artificial Intelligence and Data-Driven Approaches for Proactive Road Safety Analysis

Abstract

This systematic review analyzes 21 peer-reviewed articles (2021–2025) from ScienceDirect, Elsevier, and IEEE Xplore to examine methodological advances in road safety research. Findings reveal a paradigm shift from retrospective crash analysis to proactive, data-driven approaches, with machine learning (ML) and deep learning (DL)—particularly ensemble methods such as Random Forest, XGBoost, and neural networks—achieving crash detection accuracies of 85–92%. Explainable AI (XAI) frameworks, especially SHAP, enhance model interpretability, while hybrid and ensemble models improve predictive stability. Real-time monitoring via IoT sensors, connected vehicles, and computer vision enables surrogate safety evaluations using conflict-based metrics. Despite these advances, challenges remain regarding data heterogeneity, model transferability, privacy, and computational demands. Future directions include integrating autonomous vehicles, implementing standardized data-sharing platforms, and deploying automated safety countermeasures to transition from prediction to proactive prevention.