Geological characterization and machine learning-based mapping of the Nkout iron deposit, South Cameroon

This study investigates the petrographic, mineralogical, and structural characteristics of the Nkout Center iron deposit in southern Cameroon and integrates these observations into a machine learning–assisted geological mapping workflow. A total of 38 representative samples collected from outcrops and drill cores were analyzed using petrographic microscopy, X-ray diffraction (XRD), and X-ray fluorescence (XRF) to determine modal mineralogy and lithological variability. The dominant minerals identified include hematite, magnetite, quartz, amphibole, biotite, and subordinate accessory phases. Structural measurements focused on foliation, schistosity, and fracture orientations to characterize the tectono-metamorphic framework influencing ore distribution. To enhance geological mapping, lithological data from 195 drillholes were used to train a k-nearest neighbors (k-NN) algorithm. Preprocessing steps included categorical lithology encoding, spatial filtering, and normalization of coordinates. Multiple distance metrics Euclidean, Manhattan, and Minkowski and k-values from 1 to 15 were evaluated using five-fold stratified cross-validation, accuracy scores, confusion matrices, and ROC-AUC performance. The optimal model was obtained using k = 3 with the Euclidean distance, achieving R² = 0.91, overall accuracy = 86%, and AUC = 0.99. The final k-NN classification map delineates a wider distribution of mineralized BIF facies than previously mapped, with hematite-rich formations covering approximately 51% of the area. These results demonstrate that integrating traditional geological methods with machine learning significantly improves spatial prediction in complex, poorly exposed terrains. This approach provides a valuable framework for mineral exploration and resource evaluation in Central Africa.