Non-convex clustering using expectation maximization algorithm with rough set initialization

Abstract

An integration of a minimal spanning tree (MST) based graph-theoretic technique and expectation maximization (EM) algorithm with rough set initialization is described for non-convex clustering. EM provides the statistical model of the data and handles the associated uncertainties. Rough set theory helps in faster convergence and avoidance of the local minima problem, thereby enhancing the performance of EM. MST helps in determining non-convex clusters. Since it is applied on Gaussians rather than the original data points, time required is very low. These features are demonstrated on real life datasets. Comparison with related methods is made in terms of a cluster quality measure and computation time.