클러스터 밀도와 무관한 향상된 클러스터링 기법

In this paper, we propose a modified fuzzy clustering algorithm which can overcome the center deviation due to the Euclidean distance commonly used in fuzzy clustering. Among fuzzy clustering methods, Fuzzy C-Means (FCM) is the most well-known clustering algorithm and has been widely applied to various problems successfully. In FCM, however, cluster centers tend leaning to high density clusters because the Euclidean distance measure forces high density cluster to make more contribution to clustering result. Proposed is an enhanced algorithm which modifies the objective function of FCM by adding a center-scattering term to make centers not to be close due to the cluster density. The proposed method converges more to real centers with small number of iterations compared to FCM. All the strengths can be verified with experimental results.