superiority of one approach over another but rather reflects differences in conceptual scale and data integration philosophy. Despite the quantitative performance indicated by confusion matrices and classification metrics, predictive certainty must be interpreted cautiously. Machine learning models are inherently sensitive to input dataset selection, spatial resolution, sampling strategy, and class definition. Apparent statistical robustness may mask geological biases embedded in training data, particularly in regions where exploration density is uneven. Therefore, model outputs should not be interpreted as deterministic predictions of mineralization, but as relative indicators of permissivity consistent with the scale and assumptions of the analysis. The effectiveness of the approach ultimately depends on maintaining geological reasoning as the primary interpretative filter. Artificial Intelligence functions here as an integrative tool rather than a replacement for geological judgment. Without critical evaluation of dataset representativeness and process relevance, algorithmic outputs may reinforce existing exploration bias instead of revealing new insights. Accordingly, the results are best understood as hypothesis-generating maps that require subsequent geological validation and refinement at district and deposit scale. A further strategic extension of this framework involves the incorporation of socio-environmental and governance constraints as external decision layers applied after geological prospectivity modeling. While such variables do not form part of the Mineral System itself, they strongly influence exploration feasibility under real-world conditions. Overlaying high-permissivity geological domains with datasets representing mining conflicts, land-use restrictions, protected areas, or regulatory frameworks would allow differentiation between geological potential and operational viability. This two-stage approach preserves the integrity of the process-based Mineral Systems model while expanding its applicability toward strategic exploration planning. In regions such as southern Peru, where social license and land access represent critical exploration risks, integrating these constraints may significantly refine target prioritization and capital allocation decisions. The proposed framework should be understood as an advanced regional-scale prototype designed to support early-stage exploration targeting rather than a fully operational predictive system. While the model demonstrates internal consistency by reproducing known porphyry–skarn districts and identifying additional permissive domains, its effectiveness remains dependent on dataset resolution, proxy selection, and spatial scale. As such, the current implementation provides a robust proof-of-concept for integrating Mineral Systems with machine learning. Further validation using higher-resolution datasets and independent testing areas is required before it can be considered a decision-critical exploration tool. 4. CONCLUSIONS The Random Forest feature importance analysis indicates that Fertility is the primary permissive factor controlling regional prospectivity in the Andahuaylas–Yauri Batholith province. This is consistent with the Mineral Systems framework, where fertile magmatism represents a first-order requirement for porphyry–skarn formation at regional scale.
RkJQdWJsaXNoZXIy MTM0Mzk2