Machine learning identification of Pb–Zn deposit types using sphalerite trace-element geochemistry: Insights from a TPE-optimized SVM model and SHAP interpretation

Abstract

The unique geochemical fingerprints of trace-element distribution patterns in sphalerite are particularly useful for discriminating Pb–Zn deposit types. In this study, we developed a high-performance sphalerite classification model for deposit types based on a dataset of sphalerite analyses, using tree-structured Parzen estimator (TPE) optimization with a support vector machine (SVM) algorithm. The dataset comprises 3117 analyses of sphalerite sourced from peer-reviewed publications covering 102 representative Pb–Zn deposits worldwide spanning five major genetic types, including sedimentary exhalative (SEDEX), volcanic massive sulfide (VMS), Mississippi Valley type (MVT), skarn, and epithermal deposits. Each analysis covers 12 critical trace elements (Mn, Fe, Co, Cu, Ga, Ge, Ag, Cd, In, Sn, Sb, and Pb). The optimized model demonstrated exceptional discriminative capability. It achieved a test-set accuracy of 0.9749 and delivered consistent performance across the precision, recall, and F1-score metrics. SHAP (SHapley Additive exPlanations) interpretability analysis revealed that key indicator elements (Mn, Ge, and Co) are critical for genetic classification, although there are distinct patterns of trace elements across deposit types. Dimensionality-reduction analyses (UMAP and t-SNE) reveal distinct clustering of magmatic-hydrothermal deposits (skarn, VMS, epithermal) and sedimentary-related systems (MVT, SEDEX), reflecting systematic differences in sphalerite trace-element signatures. This methodology was validated by conducting blind, machine-learning-based classification tests on the Fule and Haobugao Pb–Zn deposits. The results suggest that the TPE-optimized SVM model can identify interpretable geochemical patterns in sphalerite, making it an effective tool for distinguishing between different types of Pb–Zn deposits.