Prediction of rail transit delays with machine learning: How to exploit open data sources

The use of public transport data has evolved rapidly over the past decades. Indeed, the availability of diverse data sources and advances in analytics have led to a greater emphasis on utilizing data to enhance public transport services. Rail transit systems have increasingly become the preferred mode of travel due to their comfort, speed, and (mostly) emission-free nature. However, persistent delays continue to be a concern. Machine learning-based prediction of transit delays is an emerging field gaining recognition. The first contribution of this paper is to illustrate how to exploit available open data to improve the prediction of rail transit delays using machine learning. Moreover, through a comparison of various well-known machine learning approaches, we show that they can yield significantly different results. Notably, the improved support vector machine method presented in this study exhibits exceptional performance and is well-suited for long-term predictions. Furthermore, we have incorporated explainable artificial intelligence techniques to identify and assess the most significant factors influencing delays. To perform experiments with the method and draw robust conclusions, three case studies featuring different rail services in major cities are provided.