Frost heave behavior of coarse-grained materials for high-speed railways subgrade: experimental analysis and machine learning predictive modeling

Coarse-grained subgrade fill has traditionally been considered frost-heave-insensitive. However, uneven frost heave in such materials has become a critical constraint for high-speed railways (HSRs) in cold regions. Short-wavelength uneven frost heave induces poor sleeper-ballast contact, leading to track geometry irregularities and localized sleeper hanging. Existing frost heave evaluation methods remain inadequate for diverse conditions. Unidirectional freezing tests were performed using a self-developed apparatus to analyze temperature evolution, moisture redistribution, and frost heave development under varying cold-end temperatures (− 2 to − 15 °C). The findings reveal that moderate temperatures (− 2 to − 5 °C) promoted water-vapor-dependent heave, while severe conditions (− 10 to − 15 °C) suppressed heave despite water supply. Combining these results with literature data, Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN) models were developed to predict frost heave rates from seven factors, with XGBoost showing superior accuracy (R² = 0.991 and RMSE = 0.092 for the test set). Variable importance analysis identifies fines content, initial water content, and water supply condition as dominant controlling factors. The study provides a useful reference for the preliminary evaluation of the frost heave behavior of HSR subgrades.