A New Fuzzy Belonging-based Multi-view K-means Clustering Algorithm

Abstract

Recently, a variety of data acquisition methods leads to the description of the same thing from different angles, which induces a large number of multi-view data. Clustering multi-view data in large-scale data analysis is of great significance and very challenging. In this paper, we present a new robust clustering algorithm, the Multi-view Fuzzy K-means clustering (MFKC) algorithm. In MFKC, a fuzzy membership method is used and optimized to describe the probability of a sample belonging to certain clusters. Compared with the hard partition based multi-view clustering algorithms, MFKC can effectively describe the fuzzy relationship between multi-view data and all clusters, and has better interpretability for the clustering results. In addition, MFKC also sets different weights for different views. By optimizing the multi-view clustering objective function, we obtain the optimal weigh of each view to reflect the importance of the view. The feasibility and effectiveness of the proposed algorithm is validated through various experimental comparisons.