The thesis addresses the problem of train occupancy, aiming to develop a system that supports the planning and optimization of railway traffic. To achieve this, we designed a predictive model to estimate passenger numbers at railway stations.
For training and testing, we used data from Slovenian Railways, which include multi-year records of passenger entries and exits, train arrival and departure times, and schedules. The dataset was further enriched with information on weather, holidays, school breaks, and other factors that can influence passenger flow.
The predictive system is based on the XGBoost method. Its performance was evaluated against two simpler models: the first predicted the average number of passengers per station, while the second considered both the station and the day of the week. Passenger numbers were found to be fairly stable, making simple models often reasonably effective. Nevertheless, our model consistently outperformed both baseline approaches.
|
