A clustering ensemble method based on cluster selection and cluster splitting

Clustering ensemble is promising to get a more robust, stable and accurate clustering result by combining multiple base partitions. In this paper, we propose a clustering ensemble method based on cluster selection and splitting. We define a novel metric to evaluate the quantity of clusters and use it to select the clusters for combination. We further define a splitting index to measure the degree of clusters to be splitted into two or more sub-clusters. Then a co-association matrix is generated from the selected and splitted clusters. Finally, the spectral clustering is performed on the matrix to get the final clustering result. The experimental results show the effectiveness of our method.