Revealing the multi-stage ore-forming history of a mineral deposit using pyrite geochemistry and machine learning-based data interpretation

Classification algorithms were constructed based on pyrite trace elements using two machine learning methods, support vector machine (SVM) and artificial neural network (ANN), to discriminate the genesis of pyrites sampled from sedimentary rock, orogenic and volcanic hosted massive sulfide (VHMS) deposits. The classifiers were trained with a dataset including published trace element compositions of 2104 pyrite samples from 92 mineral deposits or stratigraphic units. Cross validations were conducted to evaluate the performances of the classifiers on unknown samples, using a variant of the k-fold method. Each time, all pyrite samples from one of the 92 deposits/strata were set aside as the testing set, and the classifiers were trained with samples of the remaining 91. Then, the performances of the classifiers were evaluated based on whether it can correctly determine the genesis of testing deposit/stratum. The circulation was repeated 92 times for each one of the deposits/strata in the dataset, and 91 of them were correctly classified by the SVM-based classifiers, and 90 by the ANN-based. The trained classifiers were then applied to reveal the genesis of the Jiashengpan Zn-Pb deposit, which is characterized by two stages of pyrite formation: early-stage fine-grained massive and late-stage coarse-grained hydrothermal pyrite. The trained algorithms show that the early-stage is compositionally similar to sedimentary pyrite, while the late-stage has affinity to those from orogenic deposits, consistent with geological and geochemical features revealed by previous studies. This study sheds light on the power of machine learning in decoding the geochemical data of pyrite, which can well record the history of ore formation.

Automated summary

Classification algorithms based on pyrite trace elements were developed using support vector machine (SVM) and artificial neural network (ANN) to distinguish the genesis of pyrites from different types of mineral deposits. The study demonstrated the effectiveness of these algorithms in decoding the geochemical data of pyrite, showcasing the potential of machine learning in geological research.