Discriminating global ore deposit genetic types using chalcopyrite trace elements: Insights from interpretable machine learning

Chalcopyrite is the most widespread copper-bearing sulfide in Earth’s crust, the principal copper ore mineral, and an important recorder of ore-forming processes. Its trace-element composition provides valuable insights into deposit type and mineralization evolution, but the multidimensional relationships among numerous trace elements are difficult to resolve using conventional bivariate diagrams. Here, we develop an interpretable machine-learning framework for ore-deposit types classification based on chalcopyrite trace-element chemistry. A global dataset comprising 5,232 analyses from 233 deposits was compiled, covering nine genetic types across magmatic, hydrothermal, and sedimentary settings. We systematically evaluated workflows combining data transformation, class-balancing strategies, and classification algorithms. The final trained model achieved a balanced accuracy of 0.96. Feature-attribution analysis identified element-specific patterns consistent with established ore-forming processes and partitioning behavior. Nonlinear dimensionality reduction also revealed geologically meaningful genetic relationships among ore deposit types. From a comparative mineral-chemical perspective, the results also constrain trace-element incorporation in chalcopyrite under contrasting physicochemical conditions. Application of the trained machine-learning model to the Jiawula large Pb-Zn-Ag deposit in China supports an epithermal classification while indicating a genetic link with the southern porphyry stock, together defining a porphyry-epithermal system. Overall, interpretable machine learning offers a transferable framework for linking mineral chemistry, deposit classification, and metallogenic interpretation.