An Algorithm of Fuzzy Collaborative Clustering based on Kernel Competitive Agglomeration

Abstract

Kernel-based clustering generally maps the observed data to a high dimensional feature space and can usually achieve preferable classification by enlarging the difference among samples. Competitive kernel clustering creates a competitive environment by means of hierarchical method in which clusters compete for samples based on cardinalities in kernel space. Collaborative clustering implementing on several subsets can be processed by one objective function, which improves the clustering performance by sharing partition matrices among different subsets. In this paper an improved algorithm of collaborative competitive kernel clustering analysis (CCKCA) is proposed, in which the mechanism of collaboration is introduced into competitive kernel clustering. Exploiting the advantages of basic algorithms, CCKCA makes full use of the knowledge of collaborative relation among different subsets based on kernel competitive clustering. The results obtained on the benchmark datasets show that CCKCA can achieve approving clustering performance.