Feature extraction using the K-means fast learning artificial neural network

Abstract

The fast learning artificial neural network is a small neural network bearing two types of parameters, the tolerance, /spl delta/ and the vigilance, /spl mu/. By exhaustively setting the combinatorial space of these parameters, it is possible to extract the data clustering behaviour to test for significance between the obtained data clusters and the actual data. If the correlation between the clustered data output and the actual data output is high, a clustering function would likely exist in the neural network that uses the prescribed parameter set. In doing so, it is possible to extract significant factors from an array of input factors and thus determine the principal factors that contribute to the particular output. Experimental results are presented to illustrate the network's ability to extract significant factors using available test data.