Averaging refined sample centers for faster out-of-core clustering

A novel sample based clustering technique has been developed in this paper. Since traditional k-means algorithm is very time consuming for large disk resident data, sample based, out-of-core clustering techniques have gained high popularity recently. We have used the concept of elimination of measurement errors by averaging over a number of samples. Here, samples or original data set are chosen randomly, they are clustered individually, in association with a refinement technique, to produce a number of sets of refined cluster centers. Average of these refined centers are expected to form a near approximation of the true centers of the original data set. A comparison of the proposed method with some existing ones proves the efficiency and usefulness of the former.