This presentation discusses an integrated geological study aimed at improving mineral exploration within the Bektakari–Bnelikhevi ore knot, located in southern Georgia. The primary motivation for undertaking this work stems from the need to reduce exploration uncertainty in complex hydrothermal systems and to develop more effective, data-driven approaches for identifying mineralized zones. This region hosts prospective epithermal and porphyry-style mineralization associated with extensive hydrothermal alteration, but traditional exploration has often been limited by the subtle and overlapping nature of alteration signatures.

To address these challenges, we combined multiple methodologies: remote sensing analysis, detailed geochemical investigations, multivariate statistics, and machine learning. Remote sensing data, processed through FCC and CRC ratio composites, principal component analysis (PCA), and spectral index ratios, allowed us to produce refined lithological and alteration maps. Advanced supervised spectral methods—including Spectral Angle Mapper (SAM), Spectral Information Divergence (SID), and Constrained Energy Minimization (CEM)—were employed to detect and map hydrothermally altered zones such as sericitic and Na–Ca alteration, which are often closely related to structural controls.

These spatial datasets were analyzed using Google Earth Engine, QGIS, and Python-based visualization tools, facilitating the integration of spectral data with structural interpretations. In parallel, we conducted a comprehensive geochemical study involving 212 rock samples. Statistical analyses, including PCA and DBSCAN clustering, revealed two dominant alteration regimes: sulfide-rich mineralization and alkali metasomatism. Key geochemical indices such as the Alteration Index (AI) and Chlorite–Carbonate–Pyrite Index (CCPI) proved effective in delineating alteration halos, with AI values ranging from 45 to 95 and CCPI from 30 to 85. Feature importance analysis highlighted the Cu–Ag–Pb Index (32%) and Metallicity Factor (27%) as the most significant predictors of mineralized zones.

Machine learning models were then trained to classify mineralization styles, achieving high precision (above 0.85) in distinguishing epithermal and porphyry systems, though recall was lower in transitional areas (approximately 0.38), indicating complexities in overlapping alteration zones.

The results of this work demonstrate that integrating remote sensing-derived alteration maps with robust geochemical and machine learning analyses provides a powerful framework for exploration targeting. This multidisciplinary approach enables more accurate prioritization of prospective zones and reduces the uncertainty typically associated with traditional exploration methods in such complex geological settings.

Ultimately, this study offers practical strategies that can be applied in similar mineralized terrains worldwide, contributing to more efficient and cost-effective exploration programs.