Machine learning-driven classification of PbZn ore deposits using pyrite trace elements and isotopic signatures: A case study of the Gunga deposit

The mineral pyrite, commonly associated with PbZn deposits, can contain a variety of trace elements and is influenced by factors such as temperature, fluid composition, and metal source. These trace elements have long been used to differentiate between various types of PbZn deposits. However, traditional discriminant diagrams, which typically rely on two or three dimensions, fail to comprehensively capture the complex relationships between pyrite trace elements and deposit types. To address this limitation, this study employed four machine learning algorithms—random forest (RF), support vector machine (SVM), gradient boost (GB), and multilayer perceptron (MLP)—to develop classification models based on pyrite trace element compositions. The models were trained on a dataset of 5400 data points from 134 mineral deposits or stratigraphic units with trace element data from published sources. The performance of the classifiers was evaluated via cross-validation using a variant of the leave-one-group-out (LOGO) method. The study applied these machine learning models to newly obtained geochemical data from pyrite samples collected at the Gunga PbZn deposit in the Lasbela-Khuzdar metallogenic belt. The results demonstrated that the classifiers could accurately identify the source of PbZn deposits, producing reliable predictive outcomes. Specifically, the models indicated that the geochemical signature of pyrite from the Gunga deposit was derived from sedimentary-hydrothermal fluids enriched in Pb, Zn, Sb, Tl, As, and Ge, which is consistent with geological and geochemical evidence. The in-situ δ³⁴S values of pyrite ranged from −24 ‰ to +25 ‰, suggesting that the sulfur in the deposit originated primarily from coeval seawater sulfate. Additionally, Pb isotope compositions indicated crustal sources for PbZn in the Gunga deposit. The combined predictions from the classifiers, along with isotopic analyses of sulfur and lead, suggest that the Gunga PbZn deposit is a Clastic-Dominant (CD)-type deposit. These findings highlight the effectiveness of machine learning techniques in classifying ore deposits and provide new insights into the origin of the Gunga PbZn deposit.