Development of a Real-Time Crash Prediction Model for Urban Expressway

Abstract

The concept of predicting road crashes in real-time is influenced by the idea that instantaneous crash probability can be fathomed using instantaneous traffic flow data and thus road users can be informed about the existence of any hazardous traffic condition in real-time as part of proactive safety measure. This paper presents a methodology for building such a model for urban expressways using Bayesian Network. 16-month (December 2006 to March 2008) crash data and 24-hour traffic data (5 minute aggregated average speed and cumulative flow) were collected for a 2-km study section on Shinjuku 4 Tokyo Metropolitan Expressway. The model was built with 150 field crash data and later validated with 50 separate crash data taking place on the same road section. The outcome was encouraging as the newly developed model could successfully predict 74% future crashes using average crash probability as the threshold value.