Data-driven Self-Organizing Map approach for Cu–Au mineral prospectivity mapping in the Rengali Province, eastern India

Delineating prospective zones in geologically complex and data-scarce terranes requires integrating diverse geoscientific datasets through robust, objective modelling frameworks. In this study, a data-driven, unsupervised Self-Organizing Map (SOM) framework was applied to the Rengali Province (RP), eastern India—an underexplored domain between the Singhbhum Craton and the Eastern Ghats Belt, to identify prospective zones for Cu–Au mineralization. Five predictor maps derived from stream-sediment geochemistry (Cu, Zn, Au), lineament density, and aeromagnetic analytical signal were preprocessed through transformation, percentile clipping, normalization, and resampling to a uniform 1 km × 1 km grid. A 10 × 10 SOM (100 nodes) was trained on the integrated dataset (10,000 iterations; learning rate 0.5; σ = 1.0) to produce a Best Matching Unit (BMU) map, which was subsequently reclustered using K-means and evaluated with the Davies–Bouldin Index (DBI). The optimal configuration (k = 37) yielded well-separated clusters with clear geospatial coherence. High-priority clusters correspond closely with known mineralized zones, including the Adash Cu–Au prospect and mapped gossan zones along the Riamal Shear Zone, while an additional target southwest of Adash is newly identified. Comparison with a previous knowledge-driven study shows strong spatial concordance, demonstrating the capability of SOM-based analysis to objectively capture key geological patterns. This approach provides a reproducible, first-pass exploration framework for frontier terranes where limited mineral occurrence data constrain conventional modelling.