A Complete Algorithm for the Reduction of Pattern Data in the Classification of Interval Information

Abstract

The aim of this paper is to present a novel method of data sample reduction that can be applied, in particular, to the classification of interval type imprecise information. Its concept is based on the sensitivity method, inspired by artificial neural networks, while the goal is to increase the number of apposite classifications, and, consequently, to increase calculation speed. As evident in this paper, the use of reduction algorithm eliminates the particular elements of all data sample patterns which have insignificant or negative influence on the correctness of classification. The methodology was tested on pseudo-random and real data, as well as by way of comparative analysis with similar task algorithms. The presented procedure was also tested for use in situations in which the data sample representing the individual classes had been obtained by the k-means clustering procedure.