SVR achieved the best overall balance (R² = 0.7764; AUC = 0.9727; precision = 0.98; F1 = 0.6667) under a conservative threshold (0.72), while XGB showed near-perfect ranking performance (AUC = 0.9989) and full recall but lower precision. The elevated AUC values likely reflect strong predictor separability, although they may also be influenced by spatial autocorrelation and class imbalance, as random cross-validation in spatial datasets can inflate performance estimates. Prospectivity maps from SVR and XGB aligned with known porphyry systems, supporting geological plausibility. Overall, the proposed workflow demonstrates a transparent, reproducible, and scalable framework for data-driven mineral prospectivity modeling, contributing to a digital exploration architecture that reduces interpretation time, standardizes decision criteria, and enables reliable target generation aligned with faster, smarter, and responsible mineral delivery. KEY WORDS Mineral Prospectivity Mapping (MPM), Machine Learning, Copper Porphyry, Spatial Modeling, Data Integration, ASTER Remote Sensing, SVR, XGB, Geochemical Indices, Reproducible Workflow. 1. INTRODUCTION The growing global demand for critical and base metals poses significant challenges for mineral exploration and the timely delivery of new resources. While many easily accessible deposits have already been discovered, remaining resources are often deeper, subtler, or concealed beneath complex geological settings. Traditional exploration workflows typically analyze geochemical, geophysical, remote sensing, and structural datasets independently before integrating them through expert-driven interpretation. Although conceptually robust, this approach can be time-consuming, subjective, and difficult to scale efficiently across new regions and exploration programs. Mineral Prospectivity Mapping (MPM) has emerged as a systematic framework (Lou & Liu, 2023; Shaw et al., 2022; Yan et al., 2025) to integrate diverse geological datasets and identify areas with higher mineralization potential. Early MPM approaches relied heavily on heuristic weighting and linear combinations derived from expert knowledge; these methods may struggle to capture complex, non-linear interactions among geological variables and are often sensitive to subjective parameter selection. Recent advances in Machine Learning (ML) provide powerful alternatives for modeling high-dimensional geoscientific data. Algorithms such as Random Forest, Support Vector Machines, and Gradient Boosting (Lauzon & Gloaguen, 2024; Lou & Liu, 2023) have
RkJQdWJsaXNoZXIy MTM0Mzk2