Effect of Lattice Topologies and Distance Measurements in Self-Organizing Map for Better Classification

Abstract

Self-Organizing Map (SOM) is a widely used algorithm in artificial neural network for classification. Despite the general success of this algorithm, there are several limitations which some of them are poor classification accuracy and slow rates of convergence when the standard lattice topology and distance measurement are implemented. This paper investigates the performance of SOM using different topologies and different distance measurements. The results obtained showed that SOM with hexagonal topology and Euclidean distance measurement outperforms other topologies and distance measurement using at any scale datasets.