A strategic framework for road safety analysis in the connected and autonomous vehicle era

The upcoming era of connected and autonomous vehicles (CAVs) will generate vast trajectory data, enabling advanced road safety assessments. This study proposes a framework using vehicle trajectory data to identify high-risk clusters and analyse contributing factors. Using the pNEUMA dataset from Athens, Greece, comprising over 500,000 vehicle trajectories, traffic conflicts were calculated to develop a severity index based on conflict frequency, severity, and vehicle dynamics. High-risk segments were identified using a clustering algorithm and a Random Forest (RF) model with SHAP analysis evaluated contributing factors. Eleven unsafe clusters were detected, with traffic volume, conflict frequency, vehicle composition, and speed being key predictors. The RF model achieved 91% accuracy and an F1-score of 0.60. The framework offers municipalities a valuable tool to identify unsafe locations and implement targeted safety interventions, improving network-wide road safety.