Assessment of Multiple Machine Learning Models for Critical Mineral Exploration Data Quality in the Brazilian Shield

The growing integration of artificial intelligence into mineral exploration has created new opportunities for improving target selection and decision-making in geologically complex regions. This study presents an integrated multiple machine learning framework designed to address the assessment of exploration data quality. The analysis was conducted using an extensive geophysical and geochemical dataset comprising 221 exploration sites distributed across the Brazilian Shield. Six widely adopted algorithms were comparatively evaluated, including Random Forest, XGBoost, AdaBoost, Decision Trees, K-Nearest Neighbors, and Logistic Regression. The results demonstrate that Random Forest achieved the highest accuracy in data quality classification (accuracy = 0.82, AUC = 0.85). Cross-validation confirmed model robustness (5-fold CV R2 = 0.80 ± 0.02; accuracy = 0.82 ± 0.02). Feature importance and explainability analyses revealed that magnetic anomaly intensity, copper concentration, and alteration-related indices are the most influential predictors, reinforcing both the geological plausibility and the computational reliability of the models. This proposed methodology offers practical support for mineral exploration strategies across the Brazilian Shield and provides a scalable framework for future applications involving critical mineral systems.