Machine-learning applications for magmatic–hydrothermal systems: Quartz trace-element insights into ore deposits

Magmatic–hydrothermal activity is critical in ore genesis, as it forms a variety of economically significant ore deposits. Quartz chemistry is key to identifying the types and origins of such deposits; however, conventional analytical methods that use binary/ternary quartz trace-element plots are insufficient for elucidating the full complexity of quartz chemistry. To address this limitation, machine learning is applied to analyze high-dimensional quartz chemistry and improve the identification accuracy. We compiled 8710 quartz samples from a range of geological environments including various host rocks and ore deposits derived from I-, S-, and A-type magmas, and examined 18 trace elements. We developed two approaches—eXtremely Greedy tree Boosting-Recursive Feature Elimination-SHapley Additive exPlanations (XGBoost-RFE-SHAP), and Principal Component Analysis-Uniform Manifold Approximation and Projection (PCA-UMAP)—to discriminate these geological environments based on quartz trace elements. XGBoost-RFE-SHAP achieved highly accurate discrimination and identified key quartz trace elements. B, Ti, P, Li, Ge and Al are key to identifying I-, S-, and A-type parental magmas. For rocks and ore deposits related to I-type magma, key discriminators include Ti, Ge, Al, Li, Sb, and P; for S-type magma, Ge, P, Ti, Al, Li, and B; and for A-type magma, Ti, P, K, Al, Rb, and As. PCA-UMAP generated plots for classifying rocks and ore deposits, revealing quartz trace-element patterns across various geological environments. This integrated approach enhances the understanding of magmatic–hydrothermal evolution and offers a powerful tool for identifying economically viable deposits, with potential applications in diverse geological settings.