The Potential for Real-Time Traffic Crash Prediction

This paper investigates the use of real-time loop detector data in predicting crash potential. A statistical approach using a shockwave and rule-based methodology is proposed to estimate the time of the crash and to identify how much time and distance ahead of crash occurrence that loop data may be used to predict the impending hazard. Historical crashes and corresponding archived data from loop detector stations surrounding crash locations were used. Findings show that crash-prone conditions in terms of high coefficient of variation in speed and low coefficient of variation in volume are not ephemeral on freeway sections. The hazard ratio values for these variables were significantly different from ones around the crash location for several time slices (they existed for about 15 minutes), which should provide enough time for prediction and prevention of crashes. Once a potential crash location is identified in real time, measures for reducing the speed variance may be implemented to reduce the risk.