A Multi-Technique Framework For Detecting Subtle Gold Mineralisation In Obscured Terrains Of Northern Ghana

Abstract

Gold (Au) exploration in the Bole area of Ghana's Savannah Region, situated within the Birimian terrane, is hampered by severe challenges primarily due to weak surface geochemical signals and complex lithological overprints typical of greenstone belts. This study presents an integrated geochemical modelling framework to enhance gold targeting in such “low-signal” terrains. A dataset of 729 soil samples was analysed using Induc tively Coupled Plasma Mass Spectrometry (ICP-MS), with the results evaluated using a combination of Principal Component Analysis (PCA), autoencoder-based clustering, supervised machine learning classifiers (XGBoost, ANN, Random Forest), and feature sensitivity analysis. Gold distribution was found to be highly skewed, consistent with podiform or structurally controlled mineralisation. PCA isolated key components indicative of both lithological background and mineralisation vectors, most notably a strong Au–As association and a K O-Rb- Ba association interpreted as a mixed K-rich lithological and possible hydrothermal alteration signature. Autoencoder clustering revealed three latent domains, with Cluster 1 spatially aligning with elevated Au and pathfinder elements. Predictive models, especially XGBoost (accuracy = 0.82) and ANN (accuracy = 0.80), successfully classified As as a key pathfinder element, while K 2 O and Na 2 2 O were interpreted as alteration-related variables associated with hydrothermal mineralisation processes rather than direct gold indicators. These ele ments also demonstrated convergence across all methods, reinforcing their relevance to exploration. The study underscores the value of combining multivariate decomposition, non-linear feature learning, and supervised prediction to uncover subtle mineralisation signals. This approach not only improves exploration confidence in geologically complex and geochemically subdued terrains but also offers a transferable workflow for early-stage prospecting across other Birimian belts in West Africa.