Classification based on incremental fuzzy (1 + p)-means clustering

Abstract

Fuzzy clustering is often applied to determine the rules of the fuzzy rule-based classifiers (usually the antecedents only). In this work a new fuzzy clustering approach is proposed for such a purpose. The idea consists in alternating clustering of the objects from two classes with the prototypes obtained after the previous clustering not allowed to move during the current clustering. As a result each clustering provides newlocation of a single prototype. The classification quality obtained by the fuzzy rule-based classifier using the proposed clustering was compared with the Lagrangian SVM method on several benchmark databases.