71 The resulting outputs include Classification performance metrics (Figure 5) and Continuous prospectivity maps derived from both RF and ANN models (Figure 6). These maps represent relative permissivity rather than deterministic predictions of mineral occurrence. Figure 5 - Mineral prospectivity (predictive) map derived from the Random Forest model. Figure 6 - Mineral prospectivity (predictive) map derived from the Neural Network model. 2.3. Model Assumptions, Data Handling And Validation Strategy True absence of mineralization cannot be demonstrated at regional scale. Therefore, the nonmineralized class (Class 02) was defined as areas where no known porphyry–skarn occurrences are currently reported in the national geological database. This definition reflects the present state of geological knowledge rather than geological sterility. To reduce the risk of mislabeling potentially prospective but underexplored areas as negative samples, exclusion buffers were applied around known deposits prior to sampling. In addition, negative training points were spatially was balanced through controlled subsampling of negative instances. This approach allowed the models to learn discriminative geological patterns instead of simply reproducing spatial prevalence. No synthetic distributed across the study domain to avoid clustering bias and to ensure representative coverage of geological variability. Given the natural imbalance between the limited number of known deposits and the much larger area without documented mineralization, class distribution was evaluated before model training. To mitigate algorithmic bias toward the majority class, the dataset oversampling techniques were implemented, in order to preserve geological realism and avoid artificial pattern amplification.
RkJQdWJsaXNoZXIy MTM0Mzk2