Machine Learning Integration of Multi-Source Geoscience Data for Improved Mineral Prospectivity Mapping in the Nigerian Basement Complex

Conventional mineral prospectivity mapping methods in the Nigerian basement complex rely on subjective expert-driven weighting of evidence layers that inadequately captures non-linear and multi-scale spatial relationships between geological predictor variables and known mineral occurrences, leading to suboptimal exploration targeting with consequent inefficiency in mineral resource development. This study develops and applies a machine learning framework for mineral prospectivity mapping in the Nigerian basement complex, integrating aeromagnetic, radiometric, geochemical, remote sensing, and geological data for gold, lithium-bearing pegmatite, and iron ore target commodities. Regional-scale datasets covering the basement terrain from Kebbi to Borno states were compiled and harmonised onto a common spatial framework at a resolution of one kilometre. Feature engineering generated over two hundred predictor variables from the raw input data, including magnetic derivatives, spectral band ratios, geochemical anomaly scores, and structural proximity indices. Three machine learning classifiers, namely random forest, gradient boosting, and convolutional neural network architectures, were trained and validated using a curated mineral occurrence dataset derived from Geological Survey of Nigeria records and artisanal mining site surveys. Ensemble predictions from the three models were combined using a stacking approach. The study includes an original uncertainty quantification module producing confidence interval maps that communicate prediction reliability to end-users in exploration companies and government agencies. Recommendations for integration of this framework into the Nigerian Geological Survey Agency's National Mineral Assessment Programme are presented. Keywords: machine learning, mineral prospectivity, basement complex, Nigeria, random forest.