Machine learning coupled with mineral geochemistry reveals the origin of ore deposits

Abstract

As geosciences enter the era of big data, machine learning (ML) that is successful in big data, is now contributing to solving problems in the geosciences, yet there have been few applications in economic geology. This paper highlights the effectiveness of ML-based methods coupled with mineral geochemistry in revealing the origin of the Qingchengzi Pb–Zn ore field in China, which are either metamorphosed sedimentary exhalative (SEDEX) or magmatic–hydrothermal fluid related deposits. Laser ablation–inductively coupled plasma–mass spectrometry (LA–ICP–MS) pyrite trace elements coupled with decision tree (DT), K-nearest neighbors (KNN), and support vector machine (SVM) algorithms were applied to train the classification models. Testing of the DT, KNN, and SVM classifiers yielded accuracies of 98.2%, 96.4%, and 93.6%, respectively. The trained classifiers predict that the strata-bound and vein-type ore bodies at Qingchengzi ore field have a magmatic–hydrothermal origin, with DT, KNN, and SVM values of 100%, 97.4%, and 97.4%. In situ δ34S values of pyrite from strata-bound and vein-type ore bodies are 4.04‰ to 9.10‰ and 6.31‰ to 9.29‰, respectively, slightly higher than those of magmatic intrusions. In situ Pb isotopic ratios plot on the upper crust curve and yield two-stage model ages that are younger than metamorphic events in the region. Principal component (PC) analysis was used to determine the formation of the two types of mineralization. Pyrite from vein-type ore bodies (Py1) has lower contents of PC1 elements (Cu, Zn, Ge, Ag, Cd, Sn, Sb, and Pb) and higher contents of PC2 elements (Co, Ni, and Se) compared with pyrite from strata-bound ore bodies (Py2). Combined with previous fluid inclusion data, the vein-type ore bodies are inferred to have formed at higher temperatures than the strata-bound ore bodies. This study presents three visual classifiers to discriminate metamorphosed SEDEX and magmatic–hydrothermal Pb–Zn deposits. The prediction of classifiers and in situ S–Pb isotopic compositions suggest that the Qingchengzi Pb–Zn deposits have a magmatic–hydrothermal origin. The results demonstrate the effective application of ML-based methods to examine the origin of ore deposits.