Evaluation Of Clustering Techniques For Defining Stationary Domains Supported By Geostatistics

The definition of geological/geostatistical domains is the first step in building a mineral resource model. A suitable division for these domains requires some prior knowledge about the deposit geology and can be supported by a careful statistical analysis. It is crucial that data with similar characteristics are grouped together, to avoid the mixing of statistical populations, defining the so-called stationary domains. In order to assist in this definition, two unsupervised clustering algorithms were applied: Otsu and K-means. The first one is widely used in image segmentation and it is based on the exhaustive search of the data to determine the best threshold for separating them. K-means is one of the most used techniques in machine learning, and it is based on the iterative analysis of the statistical distribution. Clustering algorithms may present some shortcomings when applied to geological data, since they are based on pure statistical analysis, not considering spatial distribution or data location. Choosing the most appropriate number of domains can also be challenging. Some methods for defining the best number of groups are presented, based on the analysis of variances between/inside groups, supported by indicators variography in order to verify this definition.