Data-driven insights to reduce uncertainty from disruptive events in passenger railways

Abstract

This study investigates the predictive modeling of the impact of disruptive events on passenger railway systems, using real data from the Portuguese main operator, Comboios de Portugal. We develop models using neural networks and decision trees, using key features such as the betweenness centrality indicator, railway track, time of day, and the train service group. Conclusively, these attributes significantly predict the impact on the proposed models. The research reveals the superior performance of neural network models, such as convolutional neural networks and recurrent neural networks, in smaller data sets, while decision tree models, particularly random forest, stand out in larger data sets. The findings of this study unveil new attributes that can be employed as predictors. Additionally, they confirm, within this study's context, the effectiveness of certain traits previously recognized in the literature for mitigating the uncertainty associated with the uncertainty of the impact of disruptive events in passenger railway systems.