Survey of Classification Techniques
Steven W. Knox · Wiley series in probability and statistics · 2018
This chapter introduces a wide variety of approaches to the problem of classification, and develops an optimal classifier, the Bayes classifier, under the assumption that the joint probability distribution from which data are drawn in known. Viewing most methods as approximations of the Bayes classifier provides a common theme uniting the practical classification algorithms, which at first glance may appear to be quite different from each other. The first three methods, quadratic discriminant analysis, linear discriminant analysis, and Gaussian mixture models, are parametric methods, because they specify a parametric form for the class densities. The next two methods, kernel density estimation and histograms, are non-parametric methods and can be expensive to implement. The last method, the naive Bayes classifier, can be parametric, non-parametric, or mixture of the two. The chapter discusses the support vector machines.