Semi-Supervised Clustering Algorithm for Multi-Density and Complex Shape Dataset
Yangqiang Yu, Tianqiang Huang, Gongde Guo, Kai Li · 2008
There are many complicated data in real world, clustering analysis should be able to find the clusters of different shapes and densities. The existing typical clustering algorithms do not perform well on multi-density data. A semi-supervised clustering algorithm for multi-density dataset SCMD is proposed. The pairwise constraints: must-link and cannot-link that reflect the distribution of multi-density dataset are used. Experimental results show the algorithm can identify the clusters of varying shapes, sizes, and densities, even in the presence of noise and outliers. It is more efficient than SNN and DBSCAN.