Train delay prediction in Tunisian railway through LightGBM model

Abstract

Train delays are one of the most important problems in the railway systems across the world, which urges the development of predictive analysis-based approaches to estimate it. In fact, with the advanced big data analysis and machine learning tools and technologies, the train delay-prediction systems can process and extract useful information from the large historical train movement data collected by the railway information system. Besides, accurate prediction of train delays can help train dispatchers make decisions through timetable rescheduling and service reliability improving. We propose, in this manuscript, a machine-learning model that captures the relationship between the arrival delay of passenger trains and the various characteristics of the railway system. We also apply, for the first time, lightGBM regressor based on optimal hyper-parameters to predict train delays. To evaluate the introduced model performance, the latter is compared with that of some other widely used existing models. Its R-squared, RMSE and RME were also compared with those of Support Vector Machine, Random Forest, XGBboost and Artificial Neural Network models. Statistical comparison indicates that the LightGBM outperforms the other models and is the fastest.