Koliou P., Peithis S., Paradiso S., Yannis G., “Data-Driven Urban Road Safety Classification Integrating Telematics, Machine Learning, and Spatial Analysis”, Proceedings of the 20th Road Safety on Five Continents Conference, Leeds, UK, 3-5 September 2025.

This study employs a machine-learning approach to classify road segments in Athens based on crash risk. Data sources include OpenStreetMap (OSM), telematics data (harsh braking, acceleration, speeding), and historical crash records. The methodology is structured into three key phases: (1) data integration and preprocessing, aligning textual crash records with geospatial street networks; (2) feature engineering, incorporating traffic volume normalization to account for varying exposure levels; and (3) model training and evaluation, using an XGBoost classifier with Synthetic Minority Over-Sampling Technique (SMOTE) to address class imbalance. Performance metrics, including precision, recall, and F1-scores, ensure the model’s reliability in identifying high-risk locations. The model classified road segments as “Safe” or “Unsafe” based on telematics and crash data. Feature importance analysis identified speed variability, braking frequency, and acceleration patterns as the most significant predictors of crashes. Clustering techniques revealed high-risk urban areas, while spatial analysis pinpointed hazardous junctions. The classification model achieved strong predictive performance, with an F1-score of 0.86 for “Safe” segments and 0.80 for “Unsafe” segments. The results underscore the potential of telematics data in proactively assessing crash risks and optimizing traffic safety interventions. The findings support the integration of telematics data into urban road safety frameworks, offering a scalable solution for policymakers and traffic management authorities. The study’s methodology provides actionable insights for infrastructure planning, targeted enforcement, and real-time risk assessment. Future research could enhance model robustness by incorporating additional environmental and road condition variables, expanding coverage to diverse urban settings, and refining predictive algorithms. By leveraging data-driven safety assessments, this research contributes to evidence-based strategies for reducing crash occurrences and fostering safer urban mobility.