Ensemble machine learning model for exploration and targeting of Pb-Zn deposits in Algeria

In recent years, mineral prospectivity mapping (MPM) has been significantly advanced by the application of machine and deep learning techniques, overcoming many of the limitations inherent in traditional statistical methods. Conventional approaches often fail to capture the complex relationships between spatial patterns and mineral occurrences, lack interpretability for intricate problems, and are computationally intensive. This study seeks to enhance the understanding of metallogenic models and geodynamic factors, such as faults and thrusts, and their influence on the spatial distribution of polymetallic (Pb, Zn) deposits in Northeastern Algeria. This is achieved by integrating knowledge-driven and data-driven geological information with advanced machine learning methodologies. A multi-model ensemble framework is proposed, incorporating Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Convolutional Neural Network (CNN), and Stacking Ensemble methods. Among these, the stacking ensemble demonstrated superior performance. The model's efficacy was evaluated using a range of statistical metrics, including sensitivity, precision, F1 score, and the area under the receiver operating characteristic curve (AUC-ROC). The stacking ensemble achieved exceptional predictive accuracy, with a ROC-AUC value exceeding 98%, and demonstrated a strong capacity to predict mineralization in underexplored areas while providing robust assessments of predictive factors. Feature importance analysis underscored the critical roles of tectonic activity and metallogenic origins in influencing the occurrence of polymetallic mineralization. These findings highlight the stacking ensemble method as a highly accurate and efficient approach for mineral prospectivity mapping, offering valuable insights and a robust framework for guiding future mineral exploration initiatives.