Sedimentary exhalative (SEDEX) and volcanogenic massive sulfide (VMS) deposits represent two major classes of metallic ore systems. They commonly form in comparable geologic settings and may exhibit overlapping metallogenic signatures. Trace element compositions of pyrite and chalcopyrite can provide critical constraints for distinguishing their genetic origins. Here we compile a global geochemical dataset of pyrite and chalcopyrite and evaluate multiple machine learning (ML) classifiers, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and Stacking models. Classification accuracies reach 0.96 (RF), 0.96 (XGBoost), 0.95 (MLP), and 0.99 (Stacking) for chalcopyrite, and 0.96 (RF), 0.98 (XGBoost), 0.96 (MLP), and 0.98 (Stacking) for pyrite. Key discriminative trace elements include Ni, Pb, and Se in chalcopyrite, and Ni, Se, Pb, and Sb in pyrite, reflecting distinct enrichment mechanisms in SEDEX and VMS systems. To validate model performance, we conducted LA-ICP-MS and XRF analyses of chalcopyrite and pyrite from the Dongshengmiao Zn-Pb-Cu deposit (Inner Mongolia, China). The trained ML classifiers consistently classify fine-grained pyrite and chalcopyrite from Dongshengmiao as VMS-related. Integrated with previous petrological, geochemical, and isotopic evidence, these results suggest that submarine volcanic activity played a dominant role in ore formation. Our findings demonstrate the robustness of ML-based mineral chemistry discrimination and provide new insights into the genesis of the Dongshengmiao deposit.

Article link: https://doi.org/10.2138/am-2025-9907