predictors, increasing class separability within the study domain. Similar geological coherence between ML outputs and mineral system components has been documented in copper prospectivity mapping (Abedini et al., 2023). Nevertheless, because mineral deposits exhibit spatial continuity and clustering, random train–test splitting may partially inflate performance metrics due to spatial autocorrelation. Studies addressing imbalanced porphyry datasets emphasize the importance of spatially aware validation and imbalance correction strategies (Mantilla-Dulcey et al., 2024). Therefore, reported metrics should be interpreted as internal validation performance within the study domain rather than fully spatially independent predictive accuracy. Importantly, model validation was not based solely on statistical indicators. The spatial coherence of prospectivity maps with known intrusive bodies and alteration zones supports that predictive performance reflects geological plausibility rather than numerical overfitting, consistent with best practices in modern MPM workflows (Mitra et al., 2025; Mantilla-Dulcey et al., 2024). 5. CONCLUSION This study developed and implemented an automated, reproducible workflow for mineral prospectivity mapping in the Arequipa region of southern Peru. The methodology successfully integrated multidisciplinary datasets—geochemistry, ASTER-derived alteration indices, structural buffers, geophysics, and geological cartography—within a unified machine learning framework. SVR proved most robust for high-confidence target delineation, while XGB excelled in exhaustive anomaly detection. Their complementary strengths enable flexible application depending on exploration stage and risk tolerance. By reducing subjectivity, improving target prioritization, minimizing unnecessary drilling, and shortening pre-field evaluation cycles, this workflow supports more efficient capital allocation and contributes to accelerating responsible mineral discovery. The translation of geological concepts into quantitative spatial predictors was fundamental to predictive success. The elevated AUC values likely reflect the strong geological signal associated with porphyry systems, where intrusive centers, hydrothermal alteration halos, and structural corridors generate high predictor separability within the study domain. The spatial coherence between predicted targets and known mineralized zones supports the geological plausibility of the modeling approach. Reported performance corresponds to internal validation within the study domain, and additional spatially independent cross-validation is recommended to confirm generalization capacity. Future research should evaluate transferability to other metallogenic belts and incorporate uncertainty quantification in prospectivity outputs. Overall, this workflow represents
RkJQdWJsaXNoZXIy MTM0Mzk2