Traffic incident duration analysis and prediction models based on the survival analysis approach

In the traffic incident management process, it is important to predict the potential duration of an incident as accurately as possible. This study presents a model for estimating and predicting different duration stages of traffic incidents occurring on urban expressways, based on the initial information of the Traffic Incident Reporting and Dispatching System of Beijing. The accelerated failure time hazard-based model is used to develop the estimation and prediction models, as well as considering the unobserved heterogeneity, time-varying covariate and relationship between consecutive traffic incident duration stages. The developed models show that there are a number of different variables which affect different traffic incident duration stages. The model test results show that the developed model can generally achieve reasonable prediction results, except for with shorter or longer extreme values. The developed model will aid in traffic incident management by providing timely duration prediction based on the initial information, as soon as the traffic control centre receives the traffic incident report.