Application of local conditional autoregressive models for development of zonal crash prediction models and identification of crash risk boundaries

Abstract

ABSTRACTDeveloping conditional autoregressive (CAR) models is a common approach to address spatial autocorrelations. A main difficulty with these models is related to providing global smoothness, whereas local variations are ignored. Therefore, the main objective of the current research is to develop a zonal crash prediction model which considers localized spatial structure. Additionally, it is possible to identify the crash risks boundaries between low- and high-risk areas using spatial random effects that are locally structured (localized CAR). To judge the extent of success in achieving research goals, a case study with the collected data for Mashhad city was prepared. Also, to evaluate the performance of the proposed local CAR model, conventional models were developed and the results were compared. The results indicated that the cluster-based CAR model has the best performance. Additionally, by using the localized CAR model, about 16% of borders between adjacent units were identified as crash risk boundaries.