EXPLAINABLE MACHINE LEARNING FOR GEOPHYSICAL ANOMALY DETECTION IN MINERAL EXPLORATION

Abstract

This study presents a rigorously designed, explainable machine-learning framework for geophysical anomaly detection in mineral exploration. A comprehensive simulated dataset comprising magnetic tilt, analytic signal, Bouguer gravity gradient, and high-resolution spatial coordinates was analyzed with a Random Forest classifier to discriminate mineralized from non-mineralized sites. Data profiling confirmed complete records and a balanced class distribution, ensuring robust model development and unbiased performance evaluation. Statistical profiling revealed stable, near-Gaussian feature distributions, while correlation analysis demonstrated minimal multicollinearity, allowing each attribute to contribute independent predictive information. The Random Forest model achieved an accuracy of approximately 59 % and a receiver-operating characteristic area of 0.61, indicating moderate but geologically meaningful discrimination. Feature-importance analysis highlighted magnetic tilt and Bouguer gradient as the most influential predictors, consistent with established geophysical theory. By integrating transparent feature rankings with multiple complementary geophysical parameters, the framework provides interpretable, data-driven insights to guide exploration strategies, prioritize survey areas, and support evidence-based mineral resource assessment.