Cluster Center Initialization Method for K-means Algorithm Over Data Sets with Two Clusters

This paper defines nearest neighbor pair and puts forward four assumptions about nearest neighbor pairs, based on which a center initialization method for K-means algorithm over data sets with two clusters is build. Experiments on real data sets show that the proposed method is not preferable but at least comparable to the ones in literatures. The contribution of the proposed method is to open up a new approach to devising center initialization method for K-means algorithm.