<p>Mineral prospectivity mapping (MPM) plays a critical role in efficient resource exploration, yet traditional machine learning (ML) models face limitations due to their “black-box” nature. This paper advances MPM through an interpretable machine learning (IML) framework, integrating Shapley additive explanations (SHAP) and partial dependence plots (PDPs) to enhance transparency of data processing and mechanistic understanding of the Tongling ore district. Based on random forest (RF), support vector machine (SVM), and artificial neural network (ANN) models, this work systematically evaluated feature importance, nonlinear interactions, and spatial contributions under unified geological constraints. The results demonstrated that RF (accuracy = 0.91, AUC = 0.95) and ANN (accuracy = 0.91, AUC = 0.93) outperformed SVM in predictive performance, with intrusions, faults, and geochemical elements (e.g., Cu) identified via SHAP analysis as dominant predictors. The PDPs revealed consistent relationships between resulting prospectivity probability values and key evidential maps, such as negative correlation with distance to intrusions and positive relationship with Cu and Mo concentrations. By integrating ML interpretability and geological mechanisms, this framework provides a robust basis for data-driven MPM, offering actionable insights for targeting concealed deposits in complex terrains.</p>