Metaheuristic clustering (Q2518266): Difference between revisions

Cluster analysis means the organization of an unlabeled collection of objects into separate groups based on their similarity. Each valid group, called a ``cluster'', should consist of objects that are similar among themselves and dissimilar to objects of other groups. The aim of this volume is to investigate the application of a recently developed evolutionary computing algorithm, well known as the Differential Evolution (DE), to develop robust, fast and fully automatic clustering techniques that can circumvent the problems with several classical clustering schemes, as illustrated earlier. Since its advent in 1995, DE has drawn the attention of practitioners in optimization all over the world due to its high degrees of robustness, convergence speed, and accuracy in real parameter optimization problems. In this volume, the performance of DE is illustrated, when applied to both single and multi-objective clustering problems, where the number of clusters is not known beforehand and must be determined on the run. First, a statistical analysis of the search operators and the convergence behaviour of DE near an isolated equilibrium point in the search space is considered. A few parameter automation strategies that improve the performance of classical DE without imposing any serious additional computational burden are proposed. Next, a new DE-based crisp clustering algorithm is developed, which incorporates a new real-coded scheme for search variable representation that makes room for several possible choices of the number of clusters in the data set. The proposed clustering method is applied to an interesting problem of automatic image pixel clustering and land cover study from satellite images. The proposed clustering technique is next extended to fuzzy clustering in kernel induced feature spaces, for tackling more complex clusters, which are linearly non-separable and overlapping in nature. A new DE-variant with balanced exploration and exploitation abilities has been proposed for optimizing the clustering objectives in higher dimensional kernel space. This one perform better than the classical DE and many other recently developed algorithms for kernel-based clustering in a statistically significance fashion. Finally, the volume compares four most recently proposed multi-objective (MO) variants of the DE with two other state-of-the-art MO clustering methods over the data sets of widely varying ranges of complexity. The volume is organized in 7 Chapters, where the last one concludes with a discussion on the possible extensions of the works undertaken and projects a possible direction of future reserch.The reader is carefully navigated through the efficacies of clustering, evolutionary optimization and a hybridization of the both.