Gold mineral prospectivity mapping using gradient boosting machine and decision tree models in northwestern Ghana's Danyour Area

Gold mineral prospectivity modelling was carried out over the Danyour Area of northwestern Ghana using gradient boosting machine (GBM) and decision tree (DT) machine learning classifiers. The study responds to the escalating environmental burden of illegal artisanal mining, locally known as galamsey, whose degradation of land and water resources underscores the need for responsible, well-targeted and sustainable mineral exploration. Although earlier prospectivity mapping in the area relied on bivariate data-driven techniques that appraise each predictor independently, this study advances that work by applying multi-variate machine learning classifiers capable of resolving the non-linear interactions among geoscientific predictors that govern gold localisation. Seven geo-thematic layers derived from airborne magnetic, radiometric and geochemical datasets were integrated through the GBM and DT algorithms and trained on 218 labelled points (109 gold and 109 gold-sterile occurrences) to generate prospectivity maps for the study area. Feature-importance analysis identified arsenic concentration, the K/eTh ratio, the eU/eTh ratio and lineament density as the most influential layers, a ranking consistent with the depositional and hydrothermal processes associated with gold mineralisation in the area. Comparison of the two classifiers showed the GBM-based map to outperform the DT-based map on both discrimination and precision metrics, returning an area under the curve of 0.87 and a mean average precision of 0.916, against 0.84 and 0.806, respectively, for the DT model. Classification metrics for both models exceeded 0.7, confirming their reliability in delineating gold prospects, with 31.54% and 36.96% of the study area classified as prospective by the GBM and DT models, respectively. The predicted high-potential zones corroborate geologically favourable provinces of gold occurrence, validating the robustness of the models. These outputs provide credible exploration targets that can support land-use planning while minimising the exploration risk and the environmental and water-resource impacts associated with unregulated mining.