Prediction of Economic Potential of Lithium-bearing Rocks using Machine Learning Tools in Areas around Keffi and Akwanga, North Central Nigeria

Abstract

Resource estimation is essential for sustainable mining, mineral processing, and economic development. This study integrated remote sensing, geological mapping, and artificial neural networks (ANN) to identify, estimate, and predict the economic potential of lithium-bearing rocks in Keffi Sheet 118NE and Akwanga Sheet 119, North Central Nigeria. Landsat 8 multispectral imageries were processed, corrected, enhanced and Regolith Pattern Mapping delineated pegmatite mineralization trends, guiding geological exploration conducted in accordance with JORC standards. A total of 42 samples (35 rock and 7 soil) were analyzed using a field portable spectroradiometer and X-ray fluorescence (XRF), identifying major lithium-associated minerals including spodumene, lepidolite, quartz, albite, muscovite, and hematite. The measured Li₂O resource before ANN prediction was 2,260 g/t, with statistical analysis showing positive correlations of Li₂O with K₂O, Na₂O, and MnO, and a threshold value of 20. An ANN model (10→64→32→16→1) developed in MATLAB predicted Li₂O with an average accuracy of 76.35% and 100% confidence. Model validation produced R² = 0.99, VAF = 0.63, and RMSE = 99.7, confirming excellent agreement between predicted and measured values. The predicted lithium resource increased to 3,236.64 tons, valued at ₦1,024,937,946.50, with Angwan Doka (locations 026R and 027R) contributing 72.3% of the total economic potential. Zone A (82.4%) contained the highest lithium concentration, followed by Zones B (8.3%) and C (1.2%), demonstrating the model's effectiveness for resource estimation, mine planning, and sustainable economic development.