2316 Predictive analysis integrates geological, geochemical, and geophysical information (figure 3) to support decision-making and reduce the inherent risk of early-stage exploration. The Cajamarca region of northern Peru hosts many world-class porphyry copper and high-sulfidation epithermal deposits, making it an ideal natural laboratory to test AI-based mineral prospectivity workflows. (figure 1) 2. OBJECTIVES AND SCOPE This project pursued three main objectives: (i) to demonstrate the usefulness and capabilities of AIassisted remote-sensing interpretation and predictive analysis for the exploration and characterization of porphyry copper and high-sulfidation epithermal systems; (ii) to perform an integrated predictive analysis that combines remote-sensing interpretation, regional geology, structural data, geochemistry, and geophysics to build mineral prospectivity models; and (iii) to generate a Mineral Prospectivity Map (MPM) (figure 2) as a key tool to guide future exploration, optimize resource allocation, and provide a solid basis for continued mineral exploration in the region. The work directly addresses the WMC 2026 themes of delivering the minerals society urgently needs and getting more from what we already have by improving exploration targeting, shortening timelines to discovery, and supporting more efficient and responsible decision-making at the regional scale. 3. METHODOLOGY OR APPROACH The analysis began with extensive preprocessing of datasets from multiple sources, including geochemical measurements, geophysical surveys, and satellite imagery.(figure3) Data cleaning removed erroneous or missing values to ensure quality. Continuous variables were transformed, where appropriate, using logarithmic and exponential functions, followed by normalization to allow consistent comparison between heterogeneous datasets. Discrete variables such as lithology and faults were converted into continuous representations to enable integration into the predictive model. Fuzzy Logic and Machine Learning were selected for their ability to integrate multiple variables from different sources, including geology, geochemistry, topography, surface alteration, and slope, while combining bibliographic information with expert knowledge. Fuzzy Logic enabled manual adjustment of the relative weights of variables, incorporating hypotheses and prior experience into the design of the models, and provided flexibility to analyze related deposits and transitional zones, revealing patterns specific to the study area. The methodological workflow included data preprocessing, targeted transformations, sensitivity tests, systematic documentation of changes, and continuous validation using the F1-Score to evaluate the accuracy of the prospectivity map and balance false positives and false negatives. The core algorithm was designed specifically for the deposit types of interest, assigning relative weights to variables according to their relevance, generating fuzzy memberships that highlight areas with a high probability of hosting mineralization. The variables were then combined using Fuzzy Logic to derive the Mineral Prospectivity Map (MPM). (figure 4) The methodology also integrated passive sensors such as ASTER and WorldView with geochemical, geophysical, and topographic data. This integration helped overcome challenges such as dense vegetation and complex terrain, and improved the spatial accuracy of predictions, supporting a scalable and transferable workflow that can be applied to other metallogenic provinces. 4. RESULTS, OUTCOMES, AND PERFORMANCE The predictive analysis showed that zones with a higher number of overlapping favorable layers consistently obtained higher scores, reinforcing their prospectivity. However, isolated occurrences of certain elements or evidential layers, although associated with mineral deposits, produced some false positives in areas with limited supporting data. The model also demonstrated that individual variables can identify localized prospective areas, while Fuzzy Logic tends to emphasize deposit-related patterns in a more diffuse way, delineating zones with lower individual probability but stronger consistency across multiple key variables. In most cases, areas of high predicted probability coincided with known deposits, although some
RkJQdWJsaXNoZXIy MTM0Mzk2