Fuzzy weighted multi-kernel subspace co-clustering based on structural-α entropy

Abstract Co-clustering, which combines the rows and columns of data matrix, has achieved widespread success in many application fields. For further improving performance of clustering algorithm, we introduce the structural-α entropy of weighting vector of the co-clustering model, constructing a structural-α entropy weighting subspace clustering model, theoretically derived structural-α entropy weighting subspace two-way subspace clustering algorithm STWCC. This algorithm not only can discover clusters in different subspaces but also discover the relationship between samples and features of different subspaces in the data space, better revealing the relations between samples and features of data by maximizing the weighting entropy. On this basis, the fuzzy weighted kernel space joint clustering model of structural-α entropy is obtained by introducing fuzzy entropy of membership. By introducing the kernel method, the weighted kernel space joint clustering model of structural-α entropy is obtained. By introducing the autoencoder, the weighted subspace co-clustering model of structural-α entropy of auto-encoder is obtained, and the corresponding algorithm is derived. The proposed algorithm is experimented with synthetic datasets, and real datasets including UCI standard datasets, and is compared with some typical clustering algorithms. The results show that the proposed algorithm has a certain improvement in performance.