A novel approach to noise clustering in multivariate fuzzy c-Means
Katsuhiro Honda, Seiki Ubukata, Akira Notsu · 2017
Noise rejection is an important issue in clustering of real world data sets. In this paper, a novel robust clustering algorithm is proposed by introducing noise clustering concept into Multivariate Fuzzy c-Means (MFCM). Because MFCM is designed such that the cluster memberships of each object is estimated considering the responsibility of each attribute, the noise degree of each object is measured by introducing an additional noise attribute with a fixed (equal) value for the noise attribute of all objects. Then, the derived fuzzy memberships can work for both evaluating the cluster assignment of each object with noise rejection and evaluating the cluster-wise typical attributes. The characteristic feature of the proposed algorithm called Noise Multivariate Fuzzy c-Means (NMFCM) is demonstrated through a numerical experiment.