GEA AI: UNLOCKING HIDDEN VALUE IN COMPLEX SILVER DEPOSITS THROUGH A HYBRID GEOSTATISTICALMACHINE LEARNING FRAMEWORK POWERED BY RUST Maycol Jhonatan Benavides Sánchez Department of Mining Engineering, GEA AI Research Lab, Peru (benavidesmaycol81@gmail.com) ABSTRACT Mineral resource estimation in high-nugget epithermal deposits faces a critical dilemma: traditional geostatistical methods like Ordinary Kriging (OK) effectively model spatial autocorrelation but often smooth out high-grade variability, leading to significant underestimation of recoverable metal in selective mining scenarios. Conversely, "black-box" Machine Learning models frequently violate spatial continuity principles and lack the transparency required by international reporting codes (JORC, NI 43-101). This paper introduces GEA AI, a novel hybrid modeling architecture that integrates the spatial robustness of Kriging with the non-linear predictive capabilities of Gradient Boosting, enhanced by a high-performance geometric feature extraction engine developed in Rust. The methodology, termed "Geological Co-Stacking," operates in a multi-phase pipeline. First, a spatial baseline is established using Ordinary Kriging. Second, a parallelized Rust engine computes a 26-dimensional vector of complex geometric features for every block, including 3D Wavelet Energy to quantify textural roughness, Multifractal Detrended Fluctuation Analysis (MFDFA) to characterize local chaos, and Local Curvature to identify structural controls. Finally, an XGBoost meta-model learns to predict the residual error of the Kriging estimate based on these geometric descriptors. Validated on a complex silver deposit (CV > 3.2), GEA AI reduced the Root Mean Squared Error (RMSE) by 47.0% compared to traditional Kriging. Crucially, Grade-Tonnage analysis reveals a 55% recovery of metal content in high-grade zones (>140 g/t Ag) that traditional methods lose to smoothing. This study posits that "White Box" AI, supported by rigorous geometric auditing, represents the next standard for responsible resource estimation in the era of Mining 4.0. KEYWORDS: Resource Estimation, Machine Learning, Geostatistics, Rust, Epithermal Deposits, Multifractal Analysis, Wavelets. 1. INTRODUCTION The global mining industry faces a dual challenge: the depletion of simple, high-grade orebodies and the increasing demand for critical minerals required for the energy transition. As exploration targets become deeper and geologically more complex, traditional estimation methods often fall short. The World Mining Congress 2026 theme, "Mining For The Future," calls for a transformation in how we build trust and advance technology to meet these challenges. For decades, Ordinary Kriging (OK) has served as the industry standard "Best Linear Unbiased Estimator" (BLUE). It relies on the assumption of stationarity—that the statistical properties of the deposit (mean and covariance) are constant across the domain. However, in high-nugget epithermal veins or structurally controlled deposits, this assumption is frequently violated. The result is the well-known "smoothing effect," where the variance of the estimated grades is
RkJQdWJsaXNoZXIy MTM0Mzk2