An Information-Based Time Sequential Approach to Online Incident Duration Prediction

Abstract

Online prediction of incident duration (i.e., remaining incident duration) is becoming more important as more intelligent transportation system deployments in the United States focus on improving the operations of transportation management centers. This study proposed a time sequential procedure where an incident management process is divided into stages according to the specific information available. For each stage, a hazard-based duration regression model with different variables representing the available information was developed. By calibrating these models, the parameters of probability distributions assumed for incident duration and the coefficients of the variables were jointly estimated. Based on the estimated parameters and coefficients, remaining incident duration can be predicted online using the truncated median of incident duration. This study concluded that the accuracy of the prediction of incident duration increases as more information is incorporated into the developed models. The prediction based on the truncated median is more accurate than that based on the truncated mean because the probability distribution of incident duration has a long tail. The proposed procedure provides flexibility for implementation in the real world.