ABSTRACT Understanding the genesis of ore deposits is fundamental to resolving mineralization processes and providing a basis for effective exploration strategies. Sphalerite is among the most widely occurring Zn‐bearing sulphides and can incorporate a broad spectrum of trace elements during crystallisation, thereby serving as a sensitive geochemical proxy for reconstructing ore forming conditions. As geoscientific datasets grow increasingly complex, traditional geochemical discrimination diagrams often exhibit pronounced overlap among deposit types and account for only a limited portion of compositional variability. In contrast, machine learning approaches are capable of extracting non‐linear relationships embedded within high‐dimensional data. Here, we compile sphalerite trace element datasets from hydrothermal vein (H‐vein), porphyry and skarn Pb–Zn–Ag deposits associated with magmatic–hydrothermal systems and assess the classification performance of four algorithms, namely light gradient boosting machine (LightGBM), k nearest neighbours (KNN), convolutional neural network (CNN) and support vector machine (SVM). These algorithms represent distinct learning paradigms, enabling a systematic comparison of their predictive capabilities. Among them, LightGBM yields the highest classification accuracy and identifies Co, Fe and Ge as key discriminators that likely record variations in mineralization temperature and metal sources. Applying the optimised model to the Hua'aobaote Pb–Zn–Ag deposit in the southern Great Xing'an Range (SGXR), in conjunction with geological constraints, suggests a medium to low temperature epigenetic H‐vein origin genetically linked to a porphyry‐related magmatic hydrothermal system. These results imply that both the peripheral and deeper parts of the deposit may host additional exploration potential.
Q et al. (Sun,) studied this question.