ABSTRACT
Support vector machines (SVMs) have been applied to several areas including time series analysis, bioinformatics, textual data mining, and speech recognition. Unlike backpropagation neural networks and other techniques that often create locally optimal models, SVMs are able to provide globally optimal solutions. In addition, because SVMs use the most difficult-to-classify instances as the basis for their predictions, they are less likely to over fit the data. The basic idea of how an SVM algorithm works is conceptually simple, but the mathematics can be a challenge. To help remove much of the mystery behind SVMs, this chapter uses an example-based approach to help explain the theory and application of SVMs to real problems. This chapter concludes with a discussion of microarray data mining followed by an experiment that uses the svm function (package e1071) to build a model able to differentiate between several types of cancerous tumors observed during childhood.
