Traffic Incident Duration Prediction Based on the Bayesian Decision Tree Method

Abstract

Effective incident management requires a full understanding of various characteristics of incidents to accurately estimate incident durations and to help make more efficient decision to reduce the impact of non-recurring congestion due to these accidents. This paper presents a new prediction model base on the Bayesian decision model to estimate the traffic incident duration. The basic theory of Bayesian decision model is presented and the creation model is created based on this theory using incident data collected in Rijkswaterstaat Verkeerscentrum Nederland from various sources. Compared to most existing methods, the proposed model is unique in two aspects: firstly, the model is adaptive in the presence of real incidents for which data might only be partially available or in the presence of incomplete information. Secondly, this model is shown to perform better theoretical prediction accuracy compared to the decision model or the Bayesian model.