Poisson Family Regression Techniques for Prediction of Crash Counts using Bayesian Inference

This paper documents the application of Bayesian modelling technique for road traffic crash analysis on a sample of Indian National Highways. The study aims to identify the most critical safety influencing variables of a section of four-lane National Highway-58 through statistical models that predict frequency of accident count accurately for the provided highway safety influencing variables. The Highway traverses mainly through a plain terrain of mostly agricultural and Industrial areas. Most of the Highway study segment falls in rural areas (approximately 85%). The study has been done for newly constructed Four-Lane road between Km. 75.00 to Km. 130.00 to identify all safety deficiencies responsible for road accidents. Explanatory variables were Geometric Characteristics like Median Opening (MedOpn), Access Roads to main highway (AcsRds) and Traffic Characteristics like Average Daily Traffic (ADT) and road-side developments like Industrial (Ind), Commercial (Com), Residential (Resi) and School (School) were analyzed against dependent variable as crash count per two hundred meter per year. The results show promise towards the goal of obtaining more accurate estimates by accounting for correlations in the crash counts and over-dispersion. The results of this study show that Poisson Weibull model predicts the crashes with better accuracy. Traffic volume, Access Roads, Median opening and presence of Schools emphasized on increase in the probability of occurrence of crashes. This safety study can be useful to develop traffic safety control policies. (C) 2013 The Authors. Published by Elsevier Ltd.