Spatial analysis of road crash frequency using Bayesian models with Integrated Nested Laplace Approximation (INLA)
Romi Satria, Jonathan Agüero-Valverde, María Castro
Abstract
Improving traffic safety is a priority of most transportation agencies around the world. As part of traffic safety management strategies, efforts have focused on developing more accurate crash-frequency models and on identifying contributing factors in order to implement better countermeasures to improve traffic safety. Over time, models have increased in complexity and computational time. Bayesian models using the MCMC method have been commonly used in traffic safety analyses because of their ability to deal with complex models. Recently, the INLA approach has appeared as an alternative to the MCMC method by significantly reducing the computing time. In this study, an INLA-CAR model is developed to assess crashes by severity at the segment level on a highway section in Banda Aceh, Indonesia and is compared with a Bayesian non-spatial model. Results of the DIC show the importance of including spatial correlation in the models. The coefficient estimates show that AADT is the most influential in both models and across all severity types; however, the coefficient estimates for land use and horizontal alignment vary across severity types. Finally, in order to assess some limitations of the DIC, three other goodness-of-fit measures are used to cross-validate the results of the DIC.