Handbook of Cluster Analysis

From the wide ranging ‘Handbooks of modern statistical methods’ series, this book seeks to be a non-exhaustive guide to the subject in a large and expanding field. The book is well laid out over 31 chapters each having its own introduction and conclusion, spanning the material in a logical manner aiding accessibility. Its main focus is on partitioning sets, and care is taken to explain the exploratory nature of the analysis in contrast with the predictive task of classification (i.e. it covers unsupervised rather than supervised classification). Although the book concerns unsupervised clustering, the authors reiterate the importance of input from the researcher in suitable choice of models as ‘different algorithms often yield strikingly different clusterings from the same dataset’. Starting with a history of the field, they lay out a consistent approach to collating the different techniques from different disciplines as well as ties between fields (e.g. linkage cluster methods and ties to graph theory). This broadly moves to metric-based methods (similarity and dissimilarity measures) and then to probability models and other techniques such as semisupervised methods. The last chapters address cluster validation, whereas the longer final chapter consolidates the book by addressing the strategic choices that are needed to select an appropriate clustering technique (i.e. preprocessing data, or relating the method to clustering aims). Big data issues are explored, leading to dimension reduction and variable selection considerations such as principal component analysis or alternative aggregation techniques—then linking to the choice of suitable visualizations which is explicitly given its own chapter. Many different types of data have been used in the examples and where colour plots are crucial links from this black-and-white book to Web sites have been provided. The more recent topic of network analysis is illustrated with clear examples (e.g. a mixed membership stochastic block model), and references are given to sources of code or software, although code is not included with the examples. Cluster validation techniques such as the silhoutte criterion are explained as they become relevant (leading to how they can be extended to fuzzy clustering models) rather than linked till later in the book in the chapters on validation. This is a comprehensive reference guide, which is well organized and has an approachable style with many examples. Great care has been taken to provide an appropriate level of detail by using illustrative and topical examples. As an overview of an increasingly important field, it provides a vital first reference guide for a range of techniques and modelling considerations.