Metallogenic prediction based on geological-model driven and data-driven multisource information fusion: A case study of gold deposits in Xiong’ershan area, Henan Province, China
Mingjing Fan, Keyan Xiao, Li Sun, Yang Xu
Abstract
The traditional metallogenic prediction based on the geological model has been extensively used in prospecting, and the emerging machine learning method has also been successfully applied to mineral prospectivity prediction. In this study, we proposed a method based on the combination of model-driven and data-driven approaches, which was applied to the prospective area of the gold deposits in Xiong’ershan area. In this method, the geological prospecting model was taken as the prior knowledge of the data-driven model, and the metallogenic prediction elements of the prospecting model were constructed as the characteristic variables of the data-driven model, which improved the prediction accuracy of prospecting based on the geological model and reduced the uncertainty of outcome prediction based on data-driven. The XGBoost and LightGBM decision tree algorithms were selected for the data-driven method to create MPM. Then the SHAP framework was used to analyze the characteristic variables of the model, thus showing that geochemical prospecting could be carried out at the intersection of faults in the study area. Finally, the performance indicators based on machine learning and the legacy geological data were used to verify the robustness of the model. The AUC values of the final model were >0.95, and the new exploration targets were delineated in the high anomaly area, indicating the direction for subsequent exploration work.