Abstract Assessing the seismic risk to high-speed rail systems with limited structural damage data is a challenging task. This study employs machine learning algorithms to predict the wheel-rail forces of high-speed trains operating under post-earthquake conditions, particularly when infrastructure along the railway is damaged. By developing a post-earthquake train operation safety assessment model, this paper effectively predicts the impact of track irregularities induced by different earthquake intensities on wheel-rail interactions. The results obtained through feature selection and regression analysis using the Random Forest algorithm provide valuable insights into key safety indicators for post-earthquake train operations. Compared to traditional multibody dynamic models, this approach significantly improves the efficiency of analysis and computation. An analysis of the results across multiple speed levels reveals that trains operating at 250 km/h face derailment risks when subjected to high-intensity earthquakes (PGA > 0.6 g). The proposed predictive safety assessment model demonstrates that adjusting train speeds according to seismic intensity can effectively enhance operational safety, offering critical support for post-earthquake emergency recovery efforts.