Initialization for K-means Clustering using Voronoi Diagram

Abstract

K-Means algorithm is one of the famous partitioning clustering techniques that has been studied extensively. However, the major problem with this method that it cannot ensure the global optimum results due to the random selection of initial cluster centers. In this paper, we present a novel method that selects the initial cluster centers with the help of Voronoi diagram constructed from the given set of data points. The initial cluster centers are effectively selected from those points which lie on the boundary of higher radius Voronoi circles. As a result, the proposed method automates the selection of the initial cluster centers to supply them for K-means. The proposed method is experimented on various artificial (hand-made) as well as real world data sets of various dimensions. It is observed that it is able to produce better clustering results than the traditional K-means and the improved K-means.