Fast k-means algorithm clustering

TL;DR

This paper introduces a two-stage k-means clustering algorithm that significantly reduces computation time for large datasets by using a small subset of data for initial center estimation, then refining with the full dataset.

Contribution

A novel two-stage approach for k-means clustering that accelerates processing of large datasets by combining fast initial estimation with precise refinement.

Findings

Achieves 1-9 times speed-up on large datasets

Effective initial center estimation reduces total computation time

Maintains clustering accuracy comparable to standard k-means

Abstract

k-means has recently been recognized as one of the best algorithms for clustering unsupervised data. Since k-means depends mainly on distance calculation between all data points and the centers, the time cost will be high when the size of the dataset is large (for example more than 500millions of points). We propose a two stage algorithm to reduce the time cost of distance calculation for huge datasets. The first stage is a fast distance calculation using only a small portion of the data to produce the best possible location of the centers. The second stage is a slow distance calculation in which the initial centers used are taken from the first stage. The fast and slow stages represent the speed of the movement of the centers. In the slow stage, the whole dataset can be used to get the exact location of the centers. The time cost of the distance calculation for the fast stage is very low due to the small size of the training data chosen. The time cost of the distance calculation for the slow stage is also minimized due to small number of iterations. Different initial locations of the clusters have been used during the test of the proposed algorithms. For large datasets, experiments show that the 2-stage clustering method achieves better speed-up (1-9 times).