Stable initialization scheme for K-means clustering

Abstract

Though K-means is very popular for general clustering, its performance which generally converges to numerous local minima depends highly on initial cluster centers. In this paper a novel initialization scheme to select initial cluster centers for K-means clustering is proposed. This algorithm is based on reverse nearest neighbor (RNN) search which retrieves all points in a given data set whose nearest neighbor is a given query point. The initial cluster centers computed using this methodology are found to be very close to the desired cluster centers for iterative clustering algorithms. This procedure is applicable to clustering algorithms for continuous data. The application of proposed algorithm to K-means clustering algorithm is demonstrated. Experiment is carried out on several popular datasets and the results show the advantages of the proposed method.