Machine learning coupled with zircon trace elements revealing the diverse mineralization styles in the southern Great Xing’an range

Abstract

Zircon stands out as a crucial accessory mineral in geological studies, serving as a valuable archive of isotopic and trace element information. These characteristics make zircon a widely used indicator of magma fertility and mineralization potential. Since 2011, the southern Great Xing’an Range (SGXR) has emerged as a prominent mineralization zone, leading to numerous coeval discoveries of hydrothermal-vein, porphyry, and skarn-type deposits. In this study, we analyzed 10 ore-related granites and integrated previously published data to establish a new zircon trace element database. Using low-code machine learning solutions (PyCaret), CatBoost was identified as the best classification model for distinguishing mineralization diversity. It achieved an accuracy of 0.9217, an AUC of 0.9837, and demonstrated high recall, precision, and F1-scores of 0.9217, 0.9251 and 0.9214, respectively. The top five features for identifying mineralization, based on important scores, are Hf, T(°C), U, Eu_N/Eu_N* and Yb/Dy, indicating that variations in magmatic water content, temperature, and redox conditions play a critical role in determining ore deposit types. Quantitatively, hydrothermal-vein zircons exhibit the widest temperature range (544–945 °C; mean 758 °C) and highest Hf concentrations (mean 35,855 ppm), followed by porphyry zircons with narrower temperatures (605–954 °C; mean 752 °C) and lowest Hf contents (mean 15,860 ppm), while skarn zircons show the highest mean temperatures (597–980 °C; mean 777 °C) but the intermediate Hf contents (mean 21,788 ppm). These variations reflect differences in magmatic conditions, degrees of fractionation, and fluid–rock interaction among mineralization styles, and are further influenced by broader geological factors such as metal-enriched sedimentary strata, magmatic evolution, and tectonic setting. Integrating zircon geochemistry with regional geological context enhances our understanding of ore-forming processes and supports exploration in the SGXR.