Gold Mineral Prospectivity Mapping Using the Ensemble Model Approach, Case Study of Northwestern Thanh Hoa Province, Vietnam

Abstract

In Vietnam, gold is considered a significant mineral resource. Although gold mining activities in Thanh Hoa province contribute economically, assessing and forecasting their spatial distribution continues to pose difficulties. Mineral prospectivity mapping (MPM) is crucial for investigating, surveying, planning, and managing natural resource exploitation, including gold deposits. Recently, the machine learning models have yielded compelling results. Although many individual machine learning models have been successfully applied, challenges in MPM remain due to data limitations and the complex nonlinear relationships between existing factors and deposits. To address the above challenges, this paper presents the first application of machine learning in general, and ensemble models in particular, for building mineral prospectivity maps (MPM) in the study area. An ensemble model integrating Random Forest (RF), Support Vector Machine (SVM), and XGBoost with ten selected conditioning factors was employed to enhance predictive accuracy. The study investigates the potential of stacking ensemble learning methods for MPM using a dataset of 438 points, consisting of 219 gold placer sampling sites and 219 non-deposit sites, divided into 70% for training and 30% for testing. The prediction model results using the Receiver Operating Characteristic (ROC) curve, with Area Under the Curve (AUC) values for RF, SVM, and XGBoost and Ensemble at 0.83, 0.87, 0,81 and 0.93. Compared with three single methods, stacking ensemble had the highest AUC. The result provides a statistical approach for constructing mineral prospectivity map (12.5-meter resolution) at the regional scale using geological, geophysical, and remote sensing data.