Machine Learning-Based Prospective Modeling for Alluvial Gold Mining: A Study Area in Colombia

Abstract Alluvial gold mining has a long-standing tradition in many regions of the world and is typically conducted through sediment deposition and riverbed modification. While numerous studies have explored the application of machine learning (ML) techniques for mineral prospectivity mapping in various mining contexts, their use in alluvial gold mining environments remains limited. This study presents three ML-based approaches for prospectivity analysis in alluvial mining settings. The first is based on a natural language processing (NLP) methodology originally introduced in Australia. The second is a hybrid approach that combines a convolutional neural network with a transfer learning enhanced position encoder. The third is a linear regression model used as a baseline for comparative analysis. These models are evaluated using data from the Cauca River basin in Colombia, a region with significant alluvial gold activity. Validation results show that the hybrid neural network approaches consistently outperform both the NLP-based method and linear interpolation in this context. The proposed approach provides a data-driven, automated, and scalable methodology for resource prospecting that shows potential for application in alluvial gold mining and broader mining exploration in regions where geological mapping and manual exploration are limited or cost-prohibitive. By integrating field data with ML and spatial analysis mining companies could prioritize drill targets, reduce exploration costs, and improve the sustainability of their operations through informed decision-making.