ACCURACY OF LOCATIONAL CRASH RISK EVALUATION USING PROBE DATA: A STOCHASTIC APPROACH

Abstract

In recent years, many attempts have been made to use vehicle probe data to evaluate locational crash risk. However, there has not been sufficient theoretical discussion of the effect of using probe data on improving the accuracy of locational crash risk evaluation. In this study, at first, a theoretical model for evaluating the risk of head on crashes were developed. Next, crash data and probe data information (the number of trips and the number of sudden decelerations) at each intersection were generated by stochastic trials, and then compared the calculated theoretical crash risk with the estimated crash risk based on each of the generated data. In addition, a sensitivity analysis was conducted to examine the relativity between the theoretical frequency of crashes and the frequency of sudden deceleration. The results suggested that the effect of using the frequency of sudden decelerations from probe data may be not apparent depending on the probe vehicle penetration rate, and that there was a possibility of overestimating crash risks at safe intersections by using the frequency of sudden deceleration.