An improved clustering method based on k-means

In this paper, an improved clustering method based on k-means is proposed. The proposed method consists of two major stages split and merge stages. Initially k-means method is employed in the dataset, and in the split stage, each cluster will be split into smaller clusters with k-mean repeatedly if they are sparse. Furthermore, in the merge stage, the average distance is employed for merging standard. Experiments are tested on real and synthetic datasets. Experimental results demonstrate the proposed clustering method can detect clusters with different sizes, shapes and densities. Moreover, it outperforms the traditional k-means and single-link clustering method.