A Novel Learning Rate Decay Function of Kohonen Self-Organizing Maps Using the Exponential Decay Average Rate of Change for Image Clustering

Abstract

Clustering requires efficient selection of similarities among sample vectors and true clustering capability of the algorithm. The Kohonen Self-Organizing Maps is the most preferred unsupervised Artificial Neural Network clustering algorithm for high-dimensional or multi-dimensional data. This study introduces a new way of improving the clustering capability of the algorithm by enhancing its learning rate decay function to decrease its learning rate gradually as the training goes on through the use of the Exponential Decay Average Rate of Change. The new function allows the Enhanced Kohonen Self-Organizing Maps algorithm to converge to the minimum producing a more robust clustered datasets. The enhanced algorithm and the conventional algorithm were applied for image clustering, and the EKSOM remarkably outperformed the clustering capability of the KSOM. The introduction of EDARC function paves the way to explore the clustering and classification capability of KSOM further.