Unsupervised geochemical classification and automatic 3D mapping of the Bolshetroitskoe high-grade iron ore deposit (Belgorod Region, Russia)

We stated and solved three successive problems concerning automatization of geological mapping using the case of the Bolshetroitskoe high-grade iron ore deposit in weathered crust of Banded Iron Formation (Kursk Magnetic Anomaly, Belgorod Region, Russia). (1) Selecting a classification (clustering) method of geochemical data without reference sampling, i.e., solution of an “unsupervised clustering task”. We developed 5 rock classifications based on different principles, i.e., classification by visual description, by distribution of economic component (Fe(2)O(3)), by cluster analysis of raw data and centered log-ratio transformation of the raw data, and by artificial neural network (Kohonnen’s self-organized map). (2) Non-parametric comparison of quality of the classifications and revealing the best one. (3) Automatic 3D geological mapping in accordance with the best classification. The developed approach of automatic 3D geological mapping seems to be rather simple and plausible.