Fuzzy Separation And Shrinkage Clustering. (December 2020)

• Record Type: Journal Article
• Authors: Yu, Ning; Zong, Yanxiang; Liang, Haoran; Zhu, Heying; Liu, Kunhong; Wu, Qingqiang
• Is Part Of: Journal of physics. Volume 1693(2020)
• Publication Date: 2020-12
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1693/1/012095
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Clustering has many applications in data mining and machine learning. Fuzzy clustering methods have been widely used in clustering. However, fuzzy clustering methods still have a fatal problem: the cluster radius sensitivity problem. The cluster radius sensitivity problem means that clusters with smaller radius will predominate in clustering and obtain more data points. Aiming at this problem, we propose a fuzzy separation and shrinkage clustering algorithm (FSC). FSC uses cluster membership degrees and cluster sizes to construct a new membership distribution, and then moves the data points according to this new membership distribution. The accuracies of our algorithm on wine, iris, balance scale and seeds are as follows: 98.82%, 97.27%, 63.07% and 91.34%. Our contributions are: (1) We propose a fuzzy separation and shrinkage clustering algorithm, which can solve the cluster radius sensitivity problem. (2) The performance of our algorithm on the UCI datasets goes beyond the benchmark algorithms.