Use of the fuzzy algorithm in samples with imprecise descriptions to define typological domains.

Abstract

The identification of homogeneous spatial domains is an important step in mineral resource modeling. Inspired by the growing integration of machine learning into geostatistical modeling, this study addresses the fuzzy algorithm for automated definition of typological classes, thus improving the reliability of information. The biggest challenge in building the 3D model is optimizing the time spent grouping the domains since 53% of the samples in the database have inaccurate lithological descriptions. A combination of statistical analyzes of grades and geological characteristics was carried out to set up the fuzzy inference system based on expert knowledge. The results were verified by comparing the clustering obtained with other machine learning techniques. The algorithm proved to be effective in maintaining a better separation of typologies. Other benefits achieved using computational intelligence were the quantification of uncertainty in lithological descriptions; replicability and standardization of concepts for defining groups and automation of steps, allowing agile updating of the model when new data is inserted