ABSTRACT

Real-time crash risk prediction is important in the traffic management system aiming at improving traffic freeway safety. This paper aims to study the relationship between crash potential and lane-level data collected by single ultrasonic detectors, which were installed on the S28 freeway in Jiangsu Province, China. A binary logistic regression model is developed with multi-vehicle crash data and detector data by using the matched case-control method. The max eigenvalue is introduced to the model to utilize lane-level data sufficiently. Results show that the model with data of 0–5 minutes prior to the crash has an accuracy of 61.9%, which is quite satisfactory. The model can be used to predict real-time crash potential through the application of advanced traffic management system in China gradually.