Discriminant Analysis and Statistical Pattern Recognition

Discriminant analysis or (statistical) discrimination has proven indispensable to fields as diverse as the physical, biological and social sciences, engineering, and medicine. This comprehensive text provides perhaps the first truly modern, comprehensive and systematic account of discriminant analysis and statistical pattern recognition, with an emphasis on the fields key recent advances. With a clear look at both theoretical and practical issues, the book systematically examines each of these developments in detail. These include such new phenomena as regularized discriminant analysis and bootstrap-based assessment of the performance of a sample-based discriminant rule. Reflecting also the increasingly image-based nature of data, especially in remote sensing, the book outlines extensions of discriminant analysis motivated by problems in statistical image analysis. The sequence of chapters is clearly and logically developed, beginning with a general introduction to discriminant analysis in Chapter 1. Subsequent chapters cover likelihood-based approaches to discrimination; discrimination via normal theory-based models; distributional results for normal-based discriminant rules; practical applications of discriminant analysis; data analytic considerations with normal-based discriminant analysis; parametric discrimination via nonnormal models for feature variables; a semiparametric approach to the study of the widely used logistic model for discrimination; nonparametric approaches to discrimination, especially kernel discriminant analysis; assessing the various error rates of a sample based discriminant rule based on the same data used in its construction; selection of suitable featurevariables using a variety of criteria; and statistical analysis of image data. With dozens of illustrative tables and figures as well as over 1,200 references, the book provides a thorough and detailed examination of both the practical and theoretical aspects of the subject as well as a…