An enhanced k-means algorithm using agglomerative hierarchical clustering strategy

To overcome the drawback that the k-means algorithm is sensitive to the selection of initial centroids, we proposed an enhanced two-stage k-means algorithm. In the first stage, we begin with selecting as many as enough initial centroids, then the basic k-means algorithm is applied to get the intermediate clusters, i.e., we keep the number of initial centroids k′ large enough to eliminate the bad centroids´ effect to the result. In the second stage, the k′ intermediate clusters are merged into k result clusters using agglomerative hierarchical clustering algorithm. We have tested our algorithm on standard data sets and synthesized data set; experiments results have manifested that our algorithm can obtain higher clustering accuracy.