Classifier Design using Adaptive Resonance Theory-based Growing Self-organizing Topological Clustering

Abstract

Learning new knowledge without destroying learned one is important for classifiers. Several classifiers realize it by incorporating a clustering method. In particular, a classifier based on a growing self-organizing clustering method can autonomously and adaptively create necessary prototype nodes from training data for determining classification boundaries. However, existing approaches with the growing self-organizing clustering method suffer from excessive node creation due to an unstable learning process. This paper proposes a classifier using the adaptive resonance theory to realize stable and fast prototype node creation. Experiment results show that the proposed classifier achieves both superior classification performance and a fast learning process compared with the conventional methods.