A validation method for fuzzy clustering of gene expression data

Clustering is a key process in data mining for revealing structure and patterns in data. Fuzzy C-means (FCM) is a popular algorithm using a partitioning approach for clustering. One advantage of FCM is that it converges rapidly. In addition, using fuzzy sets to represent the degrees of cluster membership of each data point provides more information regarding relationships within the data than do alternative approaches that use crisp clustering. However, a limitation of FCM is that it requires initial specification of the number of clusters and subsequent validation of this number. Here, we propose a Bayesian method for fuzzy clustering validation using the fuzzy partition. We show that this method outperforms popular fuzzy cluster indices on both artificial and real biological datasets. Availability: The supplementary documents and the method software are at http://ouray.ucdenver.edu/~tnle/fzble.