Up to this point, a great variety of methods for regression and classification have been presented. Recall that, for regression there were the Early methods such as bin smoothers and running line smoothers, Classical methods such as kernel, spline, and nearest-neighbor methods, New Wave methods such as additive models, projection pursuit, neural networks, trees, and MARS, and Alternative methods such as ensembles, relevance vector machines, and Bayesian nonparametrics. In the classification setting, apart from treating a classification problem as a regression problem with the function assuming only values zero and one, the main techniques seen here are linear discriminant analysis, tree-based classifiers, support vector machines, and relevance vector machines. All of these methods are in addition to various versions of linear models assumed to be familiar.
KeywordsDecision Boundary Test Error Training Error Stepwise Linear Regression Single Hide Layer
Unable to display preview. Download preview PDF.