Semi-Supervised Fuzzyc-Means Algorithm by Revising Dissimilarity Between Data

We propose two approaches for semi-supervised FCM with soft pairwise constraints. One applies NERFCM to the revised dissimilarity matrix by pairwise constraints. The other applies K-FCM with a dissimilarity-based kernel function, revising the dissimilarity matrix based on whether data in the same cluster may be close to each other or the data in the different clusters may be apart from each other. Propagating given pairwise constraints to unconstrained data is done when given constraints are not sufficient to obtain the desired clustering result. Numerical examples show that the proposed algorithms are valid.