Mineral prospectivity mapping susceptibility evaluation based on interpretable ensemble learning

Abstract

In the present study, an interpretable ensemble learning-based method for mineral prediction mapping is proposed to address the limitations of traditional mineralization prediction modeling. A stacking ensemble learning model was constructed, employing random forest (RF), extreme gradient boosting (XGBoost), and AdaBoost as primary learners, and logistic regression as the secondary learner. The model’s interpretability was analyzed using local interpretable model-agnostic explanations (LIME) and shapley additive explanations (SHAP) algorithms. The lead–zinc deposits in the Changba mining area of Gansu Province served as a case study. By integrating geological and geochemical data, and selecting 18 evaluation factors, the effectiveness and interpretability of the ensemble learning model in mineralization prediction were validated. The results demonstrate that the lead–zinc prospecting map generated using the stacking model effectively correlates geological and geochemical data with known lead–zinc deposit locations, significantly enhancing the accuracy of identifying potential lead–zinc prospecting areas.