Road safety analysis aims to reduce road crashes and enhance transportation systems’ safety and efficiency. Spatial analysis of road networks plays a crucial role in evaluating and understanding these systems and how they function across different spatial localities. To that end, Artificial Intelligence (AI) tools, such as Graph Neural Networks (GNN), provide a robust framework for analyzing data structured as networks, where nodes and edges represent spatial entities. The present study exploits telematics data from naturalistic driving, which include surrogate safety measures such as harsh braking and harsh acceleration indicators, grouped by weekly time windows and integrated into a road network enriched with geometric features. This telematics-informed network is then used as input to a GNN, which generates node embeddings, which are clustered to produce weekly node-based road network partitions, revealing spatiotemporal patterns and enabling proactive identification of unsafe areas. Compared to directly clustering the raw telematics-informed network, this GNN-based approach yields better internal validation scores, improving cluster interpretability and reliability, and greater consistency across time windows. This approach enables proactive road safety management by uncovering evolving spatial risk patterns over time, supporting more targeted and detailed interventions.
| ID | pc693 |
| Presentation | |
| Full Text | |
| Tags |
