Yue, Quansheng (Southeast University), Wu, Yao (Nanjing University of Posts and Telecommunications), Lyu, Hao (Southeast University), Yuan, Quan (Tsinghua University), Guo, Yanyong (Southeast University)

Spatio-Temporal Bayesian Hierarchical Generalized Extreme Value Modeling of the Non-Stationary Traffic Conflict Extremes at Intersections

Scheduled for presentation during the Regular Session "S32a-AI-Driven Traffic Monitoring, Safety, and Anomaly Detection" (FR-LM-T32), Friday, November 21, 2025, 11:30−11:50, Southport 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

As a state-of-the-art modeling technique, extreme value theory (EVT) has been widely applied in traffic safety evaluation through traffic conflict analysis. Despite significant progress over the past two decades, challenges including spatial and temporal dependence and heterogeneity persist. To address these issues, this study develops a spatio-temporal Bayesian hierarchical modeling (BHM) approach to model non-stationary traffic conflict extremes. A series of spatio-temporal generalized extreme value (GEV) non-stationary models are proposed under different assumptions regarding spatial and temporal random effects. Specifically, the framework includes a baseline model without spatial or temporal effects, two spatial models, two temporal models, and two spatio-temporal models. To demonstrate the applicability of the proposed framework, traffic conflict data collected over the same time period from 17 intersections in the urban area of Athens are used for empirical analysis. The Bayesian approach is employed to estimate parameters of the non-stationary models. The results show that all spatial, temporal, and spatio-temporal BHM_GEV models outperform the baseline model in terms of model fit, indicating that incorporating both spatial and temporal effects substantially improves model performance. Notably, the spatio-temporal BHM_GEV models exhibit superior performance compared to models that consider only spatial or only temporal effects, underscoring the importance of jointly accounting for spatial and temporal dependencies in extreme value modeling. In addition, it is found that motorcycle rate has significant influence on the safety of intersections, highlighting the value of incorporating traffic composition into conflict-based safety evaluation.