Pattern Recognition, 4th Edition

4.18. Support Vector Machines: The Nonlinear Case

In Chapter 3, we discussed support vector machines (SVM) as an optimal design methodology of a linear classifier. Let us now assume that there exists a mappingfrom the input feature space into a k-dimensional space, where the classes can satisfactorily be separated by a hyperplane. Then, in the framework discussed in Section 4.12, the SVM method can be mobilized for the design of the hyperplane classifier in the new k-dimensional space. However, there is an elegant property in the SVM methodology that can be exploited for the development of a more general approach. This will also allow us for ...