Yield Stress Prediction of Filling Slurry Based on Rheological Experiments and Machine Learning

Cemented filling technology is an effective approach to solving tailings accumulation and goaf, with rheological properties serving as key indicators of slurry fluidity. Since slurry rheology is influenced by multiple factors, accurate prediction of its parameters is essential for optimizing filling design. In this study, we developed a model to predict static and dynamic yield stress using the extreme gradient boosting (XGBoost) algorithm, trained on 140 experimental samples (105 for training and 35 for validation, split 75:25). For comparison, adaptive boosting tree (ADBT), gradient boosting decision tree (GBDT), and random forest (RF) algorithms were also applied. Model performance was evaluated using four metrics: coefficient of determination (R2), mean absolute error (MAE), root mean square error (RMSE), and explained variance score (EVS). The Shapley additive explanation (SHAP) method was employed to interpret model outputs. The results show that XGBoost achieved superior predictive accuracy for slurry yield stress compared with other models. Analysis of importance revealed that underflow concentration had the strongest influence on predictions, followed by the binder-to-tailings ratio, while the fine-to-coarse tailings ratio contributed least. These findings highlight the potential of machine learning as a powerful tool for modeling the rheological parameters of filling slurry, offering valuable guidance for engineering applications.