An Approach based on Geographic Weighted Regression Models in Predicting Pedestrian Crashes using Exposure and Environmental Variables

Pedestrians are one of the most vulnerable users of urban roads that are directly exposed to accidents. Traffic crashes tend to be spatially dependent and is a phenomenon known as spatial dependence. The objectives of this study include identifying alternative variables for pedestrians in urban roads and identifying accident-prone areas using pedestrian exposure variables, as well as demonstrating the efficiency of spatial models in predicting pedestrian crashes. In this study, in the first step, which is the identification of exposure variables, several statistical methods have been used to identify these variables. Also, the frequency of pedestrian crashes has been predicted using six widely used models of spatial statistics, which have been evaluated based on pedestrian crash data in Tehran for the years 1396-1398. The results of this study show that the prediction of the frequency of pedestrian crashes using Poisson regression models with zero-inflated geographical weight and distribution of zero-inflated negative geographical binomials has better results based on model selection criteria than other models. This study has shown the dispersion and density of pedestrian accidents without having the volume of pedestrians and thus can be done by taking safety measures in places prone to pedestrian crashes, social costs and casualties. In this study, the use of various geographical weighted regression models to evaluate the relationship between sociological variables and crash frequencies at the area level was appropriate. Comparison of the performance of geographic weighted regression models shows the existence of significant spatial heterogeneity in the analysis.