Run-off-road Accident Prediction Model for Two-lane Highway

Abstract

A brief literature review on run-off-road (ROR) accident prediction model was made. Road geometry, traffic volume, accidents, roadside hardware and features data of 31 rural two-lane highways (total 740 kilometers) were collected to develop ROR accident prediction models. Based on four types of statistical distributions, i.e. Poisson, Negative Binomial, Zero-Inflated Poisson and Zero-Inflated Negative Binomial, ROR accident frequency, fatality and injury models were built. Elasticity analysis was made to estimate the effect of influential factors such as roadway geometry and truck traffic on these models.