Abstract: Cemented backfill technology plays a key role in the sustainable development of mines. Accurate prediction of its uniaxial compressive strength （UCS） can optimize backfill mining design and ensure the safety and stability of the stope. Traditional strength prediction methods are time-consuming and laborious. In order to explore efficient prediction methods, this paper compared four machine learning models, DeepXDE, XGBoost, LightGBM, and CatBoost based on 162 sets of test data containing seven input features （tailings, fly ash, cement, fiber content, fiber length, slurry concentration, and curing age）. The model effect was evaluated using the coefficient of determination （R²）, mean absolute error （MAE）, and root mean square error （RMSE）. The results show that the DeepXDE model has the best performance （R² = 0.959 1, MAE = 0.189 7, RMSE = 0.263 6）, which is significantly better than other comparison models. Feature importance analysis shows that cement dosage and slurry concentration are the dominant factors affecting UCS. DeepXDE can be used as an efficient prediction tool for UCS of cemented backfill materials for mines, providing an important reference for intelligent mine backfill design.