Development of a Machine Learning-Based Ore Grade Prediction System for Gold Deposits in the West African Crystalline Shield: Application to North West Nigeria

Reliable ore grade prediction from low-cost data sources is a persistent challenge in gold exploration, particularly for structurally controlled orogenic gold deposits where grade distribution is spatially erratic. This dissertation develops and validates an original machine learning-based ore grade prediction system for gold deposits in the West African Crystalline Shield, with primary application to prospects in Zamfara and Kebbi states, North West Nigeria. An ensemble of supervised machine learning algorithms, including random forest, gradient boosting, and convolutional neural networks applied to spatial data, was trained on a compiled dataset of 14,800 drill hole assay records, geophysical attributes, and structural geology features from 11 deposit analogues across the West African Shield. Transfer learning was applied to adapt the regional model to Nigerian deposit-specific training data from six prospect datasets. Model performance was evaluated by blind holdout validation and comparison with conventional kriging estimates. The developed system achieves a cross-validation root mean square error 34 percent lower than ordinary kriging for erratic high-grade intervals and demonstrates superior performance in the prediction of grade continuity in structurally complex settings. An interpretability analysis identifies structural proximity, vein density, and pathfinder element ratios as the most influential predictive features. The dissertation delivers an open-source Python implementation of the prediction framework and recommends a multi-stage prospecting workflow integrating machine learning targeting for the Nigerian gold sector. Keywords: machine learning, ore grade prediction, gold deposits, West African Shield, North West Nigeria.