Given the wide range of their major applications, during the past decade, unsupervised classification techniques (also referred to as clustering) have received special attention from many scientists, yielding new classes of algorithms, as well as refinements and improvements for existing ones. This book is a collection of 12 papers written by representative authors in the field, presenting the most recent developments in the area of partitional clustering.
Following chapter 1, which supplies a general overview of the class of model-based clustering methods, the next two chapters present more efficient variants of the standard k-means algorithm. In chapter 2, a series of strategies aiming to speed up the well-known Lloyd-Forgy algorithm are proposed, together with conclusions derived from an experimental basis concerning their performance. Because the major drawback in using the k-means algorithm resides in its sensitivity to the initial setting of cluster centers, a long series of initialization methods has been proposed, aiming to optimize the k-means algorithm from this point of view.
Chapter 3 presents an experimental comparative analysis of six linear, deterministic, and order-invariant initialization methods for the k-means algorithm. Chapter 4 covers “nonsmooth optimization-based algorithms in cluster analysis.” More recently, fuzzy concepts and tools were used to build up clustering algorithms operating with overlapping classes. Chapter 5 provides a unified framework to generalize a series of fuzzy clustering algorithms, together with a general procedure to estimate the number of clusters in a non-centralized fashion.
Two alternatives to the density-based spatial clustering of applications with noise (DBSCAN) method in density-based clustering--the black hole clustering and protoclustering algorithms--are proposed in chapter 6. The potential of the non-negative matrix factorization (NMF) method for clustering purposes is investigated in chapter 7, where the extensions sparse NMF and weakly supervised NMF are proposed in order to derive better interpretations and to allow the inclusion of extra information and user feedback in the NMF.
The problem of clustering using overlapping classes is investigated in chapter 8, where the fundamental concepts and the most frequently used overlapping partitional clustering algorithms, together with techniques to evaluate their performance, are presented in some detail. Several algorithms of the semi-supervised type for clustering, such as COP-COBWEB, COP k-means, HMRF k-means, and active fuzzy constrained clustering, are presented in chapter 9, pointing out the potential of the techniques based on partial supervision in getting better, more appropriate solutions.
A partitional algorithm based on the dissimilarity increments distribution and a consensus clustering strategy using the partitional clustering method are presented in chapter 10. Also, a validation index that can be used to evaluate several data partitions obtained by a single clustering algorithm with different initializations, parameters, and clustering ensemble methods is proposed, and a series of experimentally derived conclusions is provided in the final sections of the chapter.
The hubness effect is present in most of the applications that involve high-dimensional complex data. The developments presented in chapter 11 show that hubness can be exploited to improve clustering performance and that clustering approaches for intrinsically high-dimensional data can be based on the assumption that neighboring occurrence frequencies correlate well with local cluster centrality. The final chapter is devoted to the problem of monitoring distributed data streams based on techniques that generate clusters that evolve in time and adapt themselves to the data streams.
The content of the book is really outstanding in terms of the clarity of the discourse and the variety of well-selected examples. The enlightening comments provided by the authors at the end of each chapter and the well-chosen bibliographic references are also important features of the book.
The book brings substantial contributions to the field of partitional clustering from both the theoretical and practical points of view, with the concepts and algorithms presented in a clear and accessible way. It addresses a wide range of readers, including scientists, students, and researchers.
