Mark Last

In this paper we present a genetically enhanced version of the classical fuzzy c-means clustering algorithm. Our algorithm uses an evolutionary method to find optimal values for some scaling constants which are used to scale the various dimensions of the given data set so that clusters can be more easily detected by compensating for differences in distributions among features. We demonstrate how using un-scaled data with the conventional fuzzy c-means algorithm can lead to incorrect classification and how our algorithm overcomes the problem. We present the results of applying our method to both a synthetic data set, which we created to demonstrate the problem, and the standard Iris data set. In both cases, reduction of misclassifications was obtained by the new method, demonstrating improvement over the standard fuzzy c-means algorithm.