An improved local adaptive clustering ensemble based on link analysis

Cluster ensemble information from local adaptive clustering (LAC) contains cluster membership of each instance and subspace of each cluster, which is more precise than clustering ensemble information from K-means clustering. However, clustering ensemble information is directly derived from LAC. This paper proposes an improved local adaptive clustering ensemble (LACE) based on link analysis. Link analysis can retrieve the similarity between each pair of clusters in clustering ensemble, called WRTU. Clustering ensemble information can further be refined according to the similarity by WRTU. Therefore, the performance can further be improved. Experiments are conducted on five UCI datasets. Experimental results show that the proposed method WRTU+LACE is better than K-means clustering, LAC, Weighted LACE.