A Self-Growing and Self-Organizing Batch Map with Automatic Stopping Condition

Abstract

This paper proposes a model of self-growing and self-organizing feature map designed to alleviate the difficulty of predetermining an appropriate size and shape of the feature map suitable for the input data in the applications of the Self-Organizing Map. The proposed model progressively builds a feature map by incremental growing of the network in a way that maintains two-dimensional regular grid structure and gradual adaptation of the reference vectors by coordinated competitive learning dynamics of the Batch Map algorithm. Experimental results based on iris data set and Italian olive oil data set show that the proposed model is effective in discovering an appropriate size and shape of the network grid to manifest a suitable feature map for the input data and that the resultant feature maps are comparable to feature maps produced by the standard SOM algorithm in their quality.