Multi-source heterogeneous data‑driven strength forecasting of cement-based paste backfill using missing-data imputation and MiceForest–Bayes–XGBoost modeling

Effective management of mining-induced strata deformation depends on real-time monitoring and prediction of cement-based paste backfill (CPB) strength, particularly its unconfined compressive strength (UCS). This study develops a multi-source UCS database for cement-based paste backfill (CPB), addressing missing values with MiceForest imputation. A hybrid MiceForest–Bayes–XGBoost model with 8-fold cross-validation achieves high-precision strength prediction. SHAP analysis reveals cement–sand ratio as the most influential parameter, followed by curing time, slurry concentration and oxide composition. MiceForest outperforms other imputation methods in preserving data integrity. This interpretable framework enables accurate strength control and mix optimisation for intelligent mine backfill engineering.