REAL-TIME CRASH PREDICTION MODEL FOR URBAN EXPRESSWAY USING DYNAMIC BAYESIAN NETWORK

Abstract

Several Real-Time Crash Prediction models have been built as a tool to augment road safety since road traffic crashes are one of the world's largest public health and injury prevention problems. Crashes occurring on freeways/expressways are considered to relate closely to previous traffic conditions occurred before the crash, which are time-varying. Static Bayesian Network (SBN) model has been used in studies previously and Dynamic Bayesian Network (DBN) is a long-established extension to BNs which allow the explicit modeling of changes over time. The assumption behind the model is an event can cause another event in future but not vice-versa. Traffic is a dynamic process and time series traffic data consisting of several time intervals should be used to illustrate this dynamic process of traffic flow before crash occurrence. In this research both SBN and DBN models were built for route 4 Shinjuku Line of Tokyo Metropolitan Expressway. Twenty four DBN and 72 SBN models were built. From the six months data, 71crash and corresponding normal data were used to build the model and randomly chosen 30 crash and corresponding normal data were used for model validation process. After model building and validation, the performances of models build with BN and DBN were compared. The result shows that model built with DBN is able to predict 8.7% more crash conditions than SBN.