This paper employs an XGBoost model to predict train passenger numbers in Thailand and assess the impact of the COVID-19 pandemic on ridership. Utilizing monthly passenger data from the State Railway of Thailand (January 2015 to December 2023), the dataset is divided into pre-COVID (2015–2019) and during-to-post-COVID (2020–2023) periods. The XGBoost model, trained and tested on pre-COVID data with hyperparameters fine-tuned using grid search and cross-validation, demonstrates superior predictive performance compared to LSTM, CNN, and GRU models. Evaluation metrics, including RMSE, rRMSE, MAE, and MAPE, indicate the XGBoost model’s effectiveness in capturing ridership patterns. The model’s predictions for the during-to-post-COVID period reveal a substantial impact on train ridership, with a peak in passenger loss in April 2020 and a gradual recovery in subsequent years, although passenger numbers do not fully return to pre-pandemic levels by the end of 2023. These findings underscore the XGBoost model’s capability to accurately predict train passenger numbers and assess the pandemic’s impact, providing valuable insights for transportation planning and policy-making in Thailand. The methodology and results contribute to the understanding of the pandemic’s effects on public transportation and offer a foundation for future research in this domain.