Hierarchical dynamic modeling for highway network real-time risk forecasting with digitalized vehicle data

In traffic safety management, identifying high-risk areas prone to traffic crashes is crucial. Road authorities focus on these high-risk segments to implement strategies that mitigate the impact of recurring crashes. However, errors in identifying these hotspots can lead to inefficient resource allocation for safety improvements. These errors often stem from the reliance on aggregated traffic data for predicting crash frequency. The road traffic system is characterized by the interaction of human, vehicle and road factors and is inherently complex. While many researchers have used components of the road traffic systems in safety evaluation studies, the use of recurrent traffic patterns remains underexplored. To address this issue, this study proposes a method for hotspot identification that utilizes safety performance analysis derived from real-time traffic data and a model with various crash factors. This paper proposes a hotspot identification approach using a real-time crash prediction model for high-risk traffic patterns. Specifically, a real-time crash prediction model is developed using logistic regression to estimate the likelihood of crashes under high-risk traffic patterns. The model integrates real-time data on traffic, weather, and road geometry to estimate these probabilities. Spearman’s correlation analysis was performed to validate the proposed method. This study reveals a strong correlation between the target frequency—a measure combining crashes and hard braking events—and the number of hazardous traffic patterns identified by the real-time crash prediction model.