Modeling Crash Frequency Using Crash and Geometric Data at Freeways

Abstract

Road crashes are one of the leading causes of death and injuries in many countries around the world, which leads to enormous losses in terms of health, social, and economic aspects. Researchers are using different tools to locate, assess, and treat hazard spots in the road network. this paper aims to investigate and understand the important variables that contribute to road crashes using different crash frequency modeling techniques. Negative Binomial and Poisson regression models were used to identify the most significant variables that increase the crash frequency on the roads. The study was conducted on data obtained from the Highway Safety Information System database for a 5-years crash period in the state of North Carolina. The results of these models showed that for the Poisson model, the p-value was significant for the segment length, AADT, speed limit, right shoulder width, and median width while the left shoulder width and number of lanes weren’t significant. The coefficient estimate B sign could be used to indicate the type of contribution to the independent variable, all the dependent variables were positive signs except for speed and median width. Therefore, the increase in speed limit will decree the number of crashes. In contrast to the Poisson model, the negative binomial model showed significance only in three variables segment length, AADT, and the speed limit, the rest are not significant based on p-value, similar to Poisson the coefficient estimate B sign for the speed was negative. As expected, the increase of exposure increases the likelihood of being in a crash, therefore countermeasures are urgently needed to manage the speed, improve traffic operation, and enhance traffic safety.