SDenPeak: Semi-supervised Nonlinear Clustering Based on Density and Distance
Wenqi Fan, Chang‐Dong Wang, Jianhuang Lai · 2016
Clustering by fast search and find of Density Peaks termed DenPeak is the latest and the most popular development of unsupervised clustering that combines both density and distance. However, it suffers from significantly inaccurate performance when there is large diversity of density in different clusters in completely unsupervised. Despite a highly improved performance in semi-supervised clustering, there has been no works to incorporate supervision into DenPeak by using only a few pairwise must-link and cannot-link constraints. To address this problem, we propose a semi-supervised framework for DenPeak, namely SDenPeak, by integrating pairwise constraints to guide the clustering procedure. Experimental results confirm that our algorithm is simple but quite effective in generating satisfactory results on targeting real datasets.