An Enhanced Fuzzy Clustering with Cluster Density Immunity

Abstract

Clustering is one of the well-known unsupervised learning methods that groups data into homogeneous clusters, and has been successfully used in various applications. Fuzzy C-Means(FCM) is one of the representative methods in fuzzy clustering. In FCM, however, cluster centers tend leaning to high density area because the sum of Euclidean distances in FCM forces high density clusters to make more contribution to clustering result. In this paper, proposed is an enhanced clustering method that modified the FCM objective function with additional terms, which reduce clustering errors due to density difference among clusters. Introduced are two terms, one of which keeps the cluster centers as far away as possible and the other makes cluster centers to be located in high density regions. The proposed method converges more to real centers than FCM, which can be verified with experimental results.