Efficient global clustering using the greedy elimination method - Electronics Letters

Introduction: The K-means algorithm is used widely either as a stand-alone clustering method, or as a fast method for computing the optimal initial cluster centres for more expensive clustering methods. It employs a simple iterative scheme that performs hill climbing from initial centres, whose values are usually randomly picked from the training data. Although the algorithm is very efficient, it suffers two well-known problems: (i) the solutions are only locally optimal, and (ii) their qualities are sensitive to the initial conditions (i.e. the values of the initial centres). This Letter presents an efficient global clustering method called the Greedy Elimination Method (GEM) for alleviating these problems.