Train delays cause significant economic and operational impacts. Accurate prediction of these delays is therefore essential for improving railway service reliability and supporting effective decision-making. Thus, this chapter introduces a novel approach combining tree-based machine learning (ML) models with Conformal Prediction (CP) to estimate train delays by predicting train travel times between consecutive stations. Using real-world data from over 12.8 million production service records collected over a period of 3 years and 8 months, the proposed approach achieved a prediction accuracy of 20 s, 90% of the time. Furthermore, CP delivers statistically valid prediction intervals with an average width of 19.3 s at a 90% confidence level, successfully meeting the target coverage as demonstrated by empirical results.