Risk assessment of rear-end crashes by incorporating vehicular heterogeneity into Bayesian hierarchical extreme value models

Abstract

Extreme value theory (EVT) has been extensively used to assess road safety with traffic conflicts. However, most studies used pooled models that do not account for vehicle heterogeneity which is characterised by different static and dynamic vehicle parameters such as size, speed, acceleration, and braking capacity. This study proposes a risk assessment technique for rear-end crashes while incorporating vehicular heterogeneity. Video-based trajectory data were collected at four uncontrolled intersections, and conflicts were estimated using modified time-to-collision (MTTC). The crash risk was derived from the observed conflicts using pooled as well as Bayesian hierarchical EVT models. Unlike the pooled model, the hierarchical model revealed that crash risk varies across leader-follower pairs. Interactions that involve cars or light commercial vehicles with slow-moving vehicles are riskier. This study highlights the importance of incorporating vehicular heterogeneity in crash risk assessment. The proposed methodology can be utilised for more accurate risk assessment in heterogeneous traffic.