A New Algorithm for Optimization of K-Means Clustering with Determining Maximum Distance Between Centroids

K-means algorithm is very sensitive in initial starting points. Because of initial starting points generated randomly, K-means does not guarantee the unique clustering results so that it is very difficult to reach global optimum. A new algorithm for optimization of K- means clustering is proposed in this paper. It determines position of initial centroids in farthest accumulated distance among them. The accumulated distance metric is built at first in order to designate the initial centroids. A new initial centroid can be selected from a data which has maximum accumulated distance metric. The iterative process is needed so that the all initial centroids are determined. The new approach proposed in this paper can positionate all centroids far separately among them in the data distribution. The experimental results show effectiveness of the proposed algorithm to improve the clustering results of K-means clustering.