Predicting copper leaching from slag: an interpretable machine learning approach under oxidative sulfuric acid conditions

Abstract

Efficient recovery of copper from metallurgical waste is essential for sustainable resource utilization. This study develops an interpretable machine learning framework to predict copper leaching efficiency from copper slag under oxidative sulfuric acid conditions. A comprehensive multi-source dataset comprising 465 experimentally reported data points collected from multiple peer-reviewed studies was compiled from peer-reviewed literature. Four algorithms, Random Forest, Support Vector Regression, XGBoost, and LightGBM, were systematically optimized using 10-fold cross-validation. XGBoost demonstrated superior predictive performance with R² = 0.9794, RMSE = 3.4757, and MAE = 2.3442 on the test set. SHAP-based interpretability analysis revealed that operational parameters, particularly leaching time, acid concentration, and temperature, exert dominant influence over copper extraction, while compositional variables such as Si, S, and Al show limited direct contribution within the investigated dataset range. The nonlinear trends identified are consistent with shrinking-core kinetics and diffusion-controlled mechanisms. External validation using independent literature datasets confirmed robust generalization capability. The proposed framework provides quantitative guidance for process optimization and offers a practical tool for enhancing sustainable metal recovery from metallurgical waste.