Incident End Time Prediction During the Incident Recovery Process

Incidents can have a big impact on the train operations. By communicating prognoses of the incident end time, ProRail informs railway operators and travelers about the expected incident end time. With reliable prognoses, an overlap between activities can be achieved, leading to less delay. Currently, the prognoses during the incident are determined by the incident coordinator based on expertise. The goal of this research was to develop a prediction model which supports in-time and precise prognoses. An XGBoost prediction model is created for data-based prediction of the incident end time during the incident. From the model, features that are important for prediction are identified and the uncertainty of the predictions is determined. The model shows that the incident end time predictions during the incident converge to the actual incident duration. Based on this, it is concluded that the model is able to support in-time and precise prognoses.