Comparison of data-driven mineral prospectivity models using ensemble-learning algorithms: Implications for future gold exploration in the Neoarchean Sonakhan Greenstone Belt, Bastar Craton, central India

• Data-driven ensemble models to assess gold potential in the SGB. • A robust multi-source integration of 18 evidential datasets, with multiple data. • Random Forest > CatBoost > XGBoost > LightGBM. • The most sensitive and influential factors in prediction of gold prospective zones. • 14 new potential areas offering immediate, high-impact targets for future exploration. In recent years, machine and deep learning have significantly improved the mineral prospectivity mapping, overcoming the limitations of traditional statistical and knowledge-based methods. The conventional methods often struggle to model complex spatial relationships. They also have limited interpretability and often require high computational effort. In this study, we examined the efficacy of four data-driven ensemble-learning models, viz., random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost), for prediction of gold prospects in the Sonakhan Greenstone Belt (SGB) in Bastar Craton, central India. The alteration, geological, geophysical, and geochemical datasets were used to extricate 18 influential geo-factors. The performance metrics of the models were evaluated using receiver operating characteristic and K-fold cross-validation curves. The results indicate that the models exhibit considerable sensitivity to gold, lithology, vanadium, tungsten, ferric, and clay alterations, implying an inherent and synergistic connection system. The K-fold AUC comparison reveals that the RF, CatBoost, and XGBoost models outperform the LightGBM model, as the latter exhibits weaker stability and higher sensitivity to influential geo-factors. This study identified 14 prospective zones for gold exploration and demonstrates that these data-driven ensemble methods are highly accurate and efficient tools for mineral prospectivity mapping. The findings indicate a robust framework for gold exploration in the SGB that highlights the need for detailed geological mapping and high-resolution geophysical surveys on a periodic scale.