Petrographic, mineralogical and structural characterization of Nkout center iron deposit (south Cameroon) and geological mapping using k-NN machine learning algorithm

In the process of exploring potential resources, estimation methods are crucial instruments. The identification of lithological layers in the field of geological mapping heavily depends on how well the prediction techniques work. In this work, we use artificial intelligence specifically, machine learning and its k nearest neighbour algorithm to establish a novel paradigm of geological mapping. A petrographic and mineralogical characterization of the non-mineralized deposits was done as the initial step in this investigation. The presence of plutonic rocks such amphiboles, amphibole and biotite granites, amphibole and biotite gneisses, and amphibolite’s was disclosed by these formations. Gneisses and amphibolite, which are all rich in magnetite, are the mineralized formations. Banded iron formations, such as coarse banded BIF (CMB), hematite magnetite BIF (HMB), and magnetite hematite BIF (MHB), are subsequently formed. It was possible to ascertain the type and traits of the various structures in the research region thanks to the structural analysis, which was performed for ductile and brittle deformation. The superficial layers produced by the logging of the boreholes served as the basis for the geological mapping by k-NN. It was feasible to acquire the hyperparameters, in particular on the number of k = 3 with the Euclidian metric distance, thanks to model optimization. This model, which was used to create the final geological map, demonstrated performance with an average confusion matrix performance of 86%, a correlation coefficient R² of 91%, and an AUC for the ROC curve of 0.99.