AN INTEGRATED MACHINE LEARNING (ML) WORKFLOW IN R FOR PREDICTIVE MINERAL EXPLORATION *C. Fernández1, A. Otiniano2, J. Andrade3 1Directorate of Mineral and Energy Resources, Systematization of Economic Geological Information, INGEMMET, Lima, Peru (*Presenting author: cnfernandez@ingemmet.gob.pe ) 2Digital Innovation & Geospatial Intelligence, Earth & Environment Division, WSP Peru, Lima, Peru 3Multidisciplinary Sensing, Universal Accessibility and Machine Learning Group, National University of Engineering, Lima, Peru ABSTRACT Mining is fundamental to Peru’s economic development, but easily accessible deposits have already been discovered while deeper or subtler systems remain difficult to detect using traditional and fragmented exploration approaches. This situation widens the gap between rising global mineral demand and the industry’s ability to supply resources efficiently. To address this challenge, this study presents an automated and scalable Machine Learning (ML) workflow that integrates multidisciplinary geological datasets to identify mineral targets with greater accuracy, reduced subjectivity, shorter processing times, enhancing transparency and reproducibility in exploration decision-making while accelerating target generation, improving efficiency, and supporting faster and smarter mineral resource delivery. The methodology was applied to copper porphyry prospectivity modeling in the Arequipa region of southern Peru (17.1 x 15.5 km). Integrated datasets included ASTER-derived hydrothermal alteration indices, rock and stream-sediment geochemistry, structural fault buffers, reduction-to-the-pole (RTP) magnetic anomalies, and geological cartography. All variables were standardized and aggregated into a unified 500 x 500 m grid, generating a reproducible multidimensional dataset for predictive modeling. Seven predictive approaches were evaluated: an empirical Weighted Mean (Wmd) and six supervised ML algorithms—Random Forest (RF), Generalized Additive Model (GAM), Extreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGB), K-Nearest Neighbors (KNN), and Support Vector Regression (SVR). The objective was to predict the spatial occurrence of known deposits within the grid. Model evaluation combined statistical metrics and spatial coherence of prospectivity outputs.
RkJQdWJsaXNoZXIy MTM0Mzk2