Three-Dimensional Mineral Prospectivity Mapping by a Gradient Boosting-Based Integrated Learning Method with Data Representation Adaptability

Three-dimensional mineral prospectivity mapping (3D MPM) plays a key role in predicting deeply concealed mineral deposits; however, integrating heterogeneous datasets within machine learning frameworks remains a major source of uncertainty. In this study, we develop a gradient boosting ensemble method that explicitly adapts to different data representations and apply it to the Haopinggou gold polymetallic deposit in the western Henan metallogenic belt. Guided by mineral system theory and a 3D geological model, model performance and feature contributions are quantitatively evaluated using the SHAP framework. The results demonstrate that the binary-data-based gradient boosting model achieves higher AUC values and prediction accuracy than alternative approaches, and more effectively delineates deep exploration targets. These findings highlight the practical value of representation-aware ensemble learning for deep mineral exploration and target delineation.