Cluster analysis using learning based on connectivity relations and distribution density

The division of objects into groups based on the relationships of their connectivity and distribution density is considered. Density over a local area in the shape of a hyperball is calculated relative to its center in the sample object. The belonging of objects to disjoint classes is determined by combining local areas with specified restrictions on the distribution density values. The condition for selecting noise objects of classes from the set of boundary ones is determined. Checking the condition is necessary to adjust the composition of classes and grouping of objects in relation to their connectivity along a system of intersecting hyperballs. The identification of standards for minimum class coverage and their use in assessing the quality of the grouping are justified. To assess the shape of the group configuration, an indicator is proposed, the value of which is the length of the minimum shortest path between the standards.

Hali tarjima qilinmagan