Application of explainable machine learning to characterizing apatite fertility in porphyry-skarn deposits

Le Wang, Ben Qin, Massimo Chiaradia, Ke-Zhang Qin, Ming-Jian Cao,
Application of explainable machine learning to characterizing apatite fertility in porphyry-skarn deposits,
Ore Geology Reviews,
Volume 186,
2025,
106926,
ISSN 0169-1368,
https://doi.org/10.1016/j.oregeorev.2025.106926.
(https://www.sciencedirect.com/science/article/pii/S016913682500486X)
Abstract: Over the past century, numerous near-surface porphyry-skarn deposits have been discovered, mined, and depleted, driving the need for innovative exploration methods to support a green-energy-driven society. Whole-rock analyses blend signals from diverse minerals, while zircon, forming late in magma evolution, lacks information on magmatic volatiles critical for mineralization. This study introduces a novel method using global igneous apatite compositional data, analyzed with supervised machine learning algorithms—Extreme Gradient Boosting (XGBoost) and Random Forest (RF). Optimized via grid search cross-validation, these models achieved classification accuracies of 94 % (XGBoost) and 89 % (RF), with feature importance analysis identifying Mn, Sr, Cl, La, REE, F/Cl, Ce/Ce*, and Eu/Eu* as key indicators distinguishing fertile from barren apatites. To validate practical utility, the models were tested on apatites from barren host granitic rocks and fertile dacite porphyry at the Güzelyayla porphyry Cu-Mo deposit in Turkey. XGBoost and RF attained accuracies of 92 % and 98 %, demonstrating their ability to serve as reliable tools for early-stage exploration. By integrating explainable machine learning with host rock geochemistry, the study links apatite fertility indicators to critical formation factors of porphyry-skarn deposits, offering a robust framework for identifying new resources in greenfield areas. Our findings demonstrate that fertile magmas possess higher Cl concentrations than barren magmas under equivalent extents of sulfide saturation and oxygen fugacity. Additionally, high oxygen fugacity and extensive fractionation favors the formation of porphyry-skarn Cu/Mo/Au mineralization.
Keywords: Porphyry-skarn deposits; Igneous apatite; Mineralization potential evaluation; Machine learning