Railway Timetable Forecasting Based on Feature Engineering and Transformer

Accurate prediction of train delays is critical for improving the efficiency of railway networks. However, traditional models often suffer from limited feature representation and inadequate handling of complex temporal patterns. To address these challenges, a Transformer-based model combined with feature engineering is proposed to enhance predictive performance. The model is evaluated using multiple metrics, including MAE, RMSE, R² and prediction accuracy within defined error tolerance thresholds. Experimental results demonstrate that the proposed approach significantly outperforms models without feature selection, benefiting from the elimination of irrelevant features and the Transformer's ability to capture long-term dependencies. Although large-scale validation across the UK railway network remains a topic for future research, the model exhibits strong scalability and practical applicability, offering a robust solution for train operation scheduling and delay management.