On Two Variants of Fuzzy Clustering for Relational Data
Shujin Saga, Yuchi Kanzawa · 2024
In this paper, two fuzzy clustering algorithms are proposed for relational data. The first algorithm is obtained by regularizing a conventional algorithm for relational data, which is similar to how the Yang's penalized Fuzzy c-means can be regarded as regularizing the Bezdek's Fuzzy c-means. The second algorithm is obtained using the following two steps: First, separating the fuzzification parameter for membership and the cluster size controller in a conventional algorithm for relational data, second, regularizing the Tsallis-entropy. This procedure is similar to that a conventional algorithm for vectorial data can be regarded as separating the fuzzification parameter for the membership and that for the cluster size controller in that for another conventional algorithm and then regularizing the Tsallis-entropy. The features of our proposed algorithms are observed through numerical experiments using an artificial dataset.