Machine learning-based classification of geological structures from magnetic anomaly data: Case study of Northern Nigeria basement complex

• Machine learning achieves 95.5% accuracy in mineral structure prediction using geomagnetic data. • Deep-to-shallow (280-2000 m) mapping reveals new mineralization zones with strong correlation (r = 0.956). • Random Forest analysis identifies strategic mineral deposits across Het-Kagoro-Durbi regions. • Methodology successfully validates mineralized structures in complex terrains. The geological terrain of Northern Nigeria presents a complex mineral resource landscape that requires systematic exploration. This study applies a machine learning framework to geomagnetic data to enhance the identification of subsurface mineralized structures. Through the integration of analytic signal processing with machine learning classifiers (Random Forest (RF) and Gradient Boosting (GB)), we analyze magnetic anomalies to predict subsurface geological features with a classification accuracy of 95.5%. The results identify mineral-rich zones across various depths, ranging from near-surface (280 m) to deep crustal levels (> 2000 m), with key prospective areas including Het, Kagoro, and Durbi. These regions contain mineral deposits such as monazite, tantalite, columbite, tourmaline, beryl, and kaolin. The study achieves a Pearson correlation coefficient of 0.956 between predicted and observed subsurface structures, demonstrating the effectiveness of this approach in mineral exploration. The methodology not only validates known geological features but also reveals previously unrecognized mineral-rich structures, contributing to a more data-driven strategy for resource assessment in geologically complex regions.