Discrimination of deposit types using magnetite geochemistry based on machine learning

The application of trace elements in magnetite for deposit genesis research is significant, highlighting its potential as a valuable mineral exploration tool. However, traditional low-dimensional analysis methods are not effective in revealing the genetic types of deposits using magnetite trace elements, as they fail to fully utilize the rich high-dimensional information provided by magnetite trace element analysis. To address this limitation, we implemented a supervised machine learning method, eXtreme Gradient Boosting (XGBoost), to correlate the multi-element composition of magnetite with deposit types. Our study encompassed 3,865 magnetite trace element datasets from six distinct deposit types (BIF, IOA, IOCG, magmatic, porphyry, and skarn deposits). It demonstrates the XGBoost classifier's efficiency and accuracy in classifying high-dimensional magnetite trace element data based on deposit types, achieving an impressive overall accuracy of 96% with an F1 score of 95%. Interpretation of the model using the SHAPley Additive exPlanations (SHAP) tool shows that Ni, Ga, Sc, and V are the most indicative elements for classifying deposit types using magnetite trace element chemistry. Additionally, a visualization method based on XGBoost and t-SNE was proposed. Finally, Xgboost and SHAP were applied in Jinchuan magmatic sulfide Ni-Cu-PGE deposit. Based on optical characteristics and machine learning, Jinchuan magnetite can be classified into three types. Compared with type I magmatic magnetite and type III hydrothermal magnetite, type II magnetite's paragenetic with sulfides, calcite and apatite and high Ni content may indicate that it's origin of interaction between sulfide melts–volatile fluids. These results indicate that deep volatile fluid plays a key role for the formation of Jinchuan Cu-Ni sulfide deposit.