Classification of lithotypes in an iron deposit based on physical and chemical properties using machine learning

This study uses machine learning techniques to address the classification of lithological units in an iron deposit based on their physical and chemical properties. The importance of this problem lies in the challenges of identifying the material origin when traceability is lost during processing. Supervised learning algorithms, including Random Forest, were applied to data from 18 boreholes comprising multiple lithotypes. Initially, the model achieved an f1-score of 0.89 for lithology classification. By grouping similar lithologies, notably different types of itabirites, the score improved to 0.97. These results suggest introducing new features, such as compactness, could further refine accuracy. This work demonstrates that machine learning offers a promising methodology for lithotype classification and may be adapted to other geological contexts.