Comparative assessment of random forest and support vector machine in gold vein prediction

Abstract

Accurate forecasting of ore veins is essential in mining due to the intrinsic variability in vein geometry, grade, and orientation, often complicated by the nugget effect which challenges conventional resource estimation. Traditional geostatistical methods like kriging frequently struggle with the complex geometries and skewed grade distributions typical of vein deposits. This research evaluates the performance of machine learning algorithms, specifically Random Forest (RF) and Support Vector Machines (SVM), for predicting the presence and continuity of gold veins in the Tiira Gold Fields, Uganda. Utilizing geological survey data (gold concentration) combined with geophysical measurements (magnetic intensity, resistivity, chargeability) and geospatial coordinates, predictive models were developed following rigorous data integration, cleaning (using Inverse Distance Weighting for interpolation), and preprocessing. A comparative assessment of RF and SVM was conducted for both binary classification (vein presence > 10 ppm) and regression (gold concentration prediction). Results indicated strong performance for both models in classification, with RF achieving a higher Area Under the Curve (AUC) (0.953) but SVM showing better precision (0.909) on test data. For regression, SVM demonstrated superior performance in predicting gold concentration (Test R² = 0.718) compared to RF (Test R² = 0.617). Sensitivity analysis identified chargeability and resistivity as the most influential predictors. This study provides a robust framework for applying machine learning in mineral exploration within complex geological settings, aiding informed decision-making and potentially reducing unnecessary excavation, thereby supporting sustainable resource management.