An evolutionary fuzzy c-means approach for clustering of bio-informatics databases

Abstract

Recently, the scientific community has started to show increasing interest in finding clusters in high-dimensional data sets such as gene product (protein or RNA) data sets in bio-informatics. In this paper we consider the problem of finding fuzzy clusters in such very high dimensional data. In fact, even if fuzzy clustering has been successfully applied to numerous data sets, for such high-dimensional databases it often produces trivial solutions where all cluster centers coincide and all memberships are equal. To solve this problem, we present an evolutionary approach that integrates fuzzy c-means clustering and feature selection. Reducing the dimensionality of the space, feature selection improves the quality of the partitions generated, and, at the same time, can help to build both faster and more cost-effective predictors, as well as a better understanding of the underlying generation process. We exhibit the good quality of the clustering results by applying our approach to two real-world data sets from bio-informatics.