Investigating the transferability of Bayesian hierarchical extreme value model for traffic conflict-based crash estimation

Abstract

The use of the extreme value theory to estimate crashes from traffic conflicts has been gaining popularity in road safety analysis. A recent advancement is the development of Bayesian hierarchical extreme value models (BHEVM) which can combine conflict extremes of different sites and account for non-stationarity and unobserved heterogeneity for crash estimation. This paper investigated the transferability of BHEVM developed based on actual vehicle trajectory data collected from the city of Surrey, Canada to two corridors of signalized intersections in Los Angeles and Georgia, USA. Two approaches were used to transfer the models: (i) through the recalibration of the random error terms and (ii) using informative priors. The results show that the Surrey model is more transferable to Georgia than to Los Angeles, and using informative priors significantly improves the transferability. The results suggest that the BHEVM is transferable if there are similarities in the base and application contexts.