Improved algorithm of cluster analysis with the application of potential codes

Abstract

There had been suggested the algorithm for clusterization of large size of data sample, the description of which can be presented in different features spaces, with the application of potential codes. Its derivation is based on the idea of cluster analysis according to method of dynamic kernels. An algorithm foresees preliminary determination of clusters centres and the formation of kernels from a few selective points, with further forming of complete cluster by the search of kernels set and attributing to them the sample vectors in a way, which allows to get clusters minimizing the criterion of fitting of distances and their grades between the points of space. An algorithm is tested on a standard iris data file.