Semi-supervised kernel-based fuzzy C-means with pairwise constraints
Na Wang, Xia Li, Xuehui Luo · 2008
Clustering with constraints is an active area in machine learning and data mining. In this paper, a semi-supervised kernel-based fuzzy C-means algorithm called PCKFCM is proposed which incorporates both semi-supervised learning technique and the kernel method into traditional fuzzy clustering algorithm. The clustering is achieved by minimizing a carefully designed objective function. A kernel-based fuzzy term defined by the violation of constraints is included. The proposed PCKFCM is compared with other clustering techniques on benchmark and the experimental results convince that effective use of constraints improves the performance of kernel-based clustering. As for the effect of key parameter selection and the non-linear capability, it outperforms a similar semi-supervised fuzzy clustering approach Pairwise Constrained Competitive Agglomeration (PCCA).