Kernel Based Fuzzy Ant Clustering with Partition Validity

We introduce a new swarm intelligence based algorithm for data clustering with a kernel-induced distance metric. Previously a swarm based approach using artificial ants to optimize the fuzzy c-means (FCM) criterion using the Euclidean distance was developed. However, FCM is not suitable for clusters which are not hyper-spherical and FCM requires the number of cluster centers be known in advance. The swarm based algorithm determines the number of cluster centers of the input data by using a modification to the fuzzy cluster validity metric proposed by Xie and Beni. The partition validity metric was developed based on the kernelized distance measure. Experiments were done with three data sets; the Iris data, an artificially generated data set and a Magnetic Resonance brain image data set. The results show how effectively the kernelized version of validity metric with a fuzzy ant method finds the number of clusters in the data and that it can be used to partition the data.