Strength prediction of cemented paste backfill with different machine learning and SHapley Additive exPlanation (SHAP) approaches

The rapid expansion of the mining industry has led to the accumulation of tailings, posing significant environmental challenges. One effective strategy to address this issue is the utilization of tailings in cemented paste backfill (CPB). However, traditional experimental methods for evaluating the unconfined compressive strength (UCS) of CPB are time-consuming and cost-intensive. To address these limitations, this study employs the Optuna-TabPFN model, which leverages Optuna for hyperparameter optimization of the tabular prior-data fitted network (TabPFN) algorithm, to accurately predict the UCS of CPB based on a small dataset of 120 UCS test samples. The proposed model achieved a high prediction accuracy, with a determination coefficient (R²) of 0.98, a mean square error (MSE) of 0.03 MPa 2 , a mean absolute error (MAE) of 0.11 MPa,and performance index (PI) of 1.80 on the testing set. Furthermore, SHapely Additive exPlanation (SHAP) analysis was employed to further interpret the model predictions and identify key factors influencing CPB strength. The analysis revealed that the tailings-to-cement ratio, curing age, and backfill concentration were the most critical factors affecting CPB strength. Additionally, a user-friendly visualization interface was developed to enhance the model's practicality and ease of use. This study provides significant theoretical support and practical tools for the design and application of CPB, especially in scenarios with limited data.