Enhancing Kernel Methods for Pattern Classification

Kernel methods are a new family of techniques §with sound theoretical grounds. They §have been shown to be powerful approaches to pattern §classification problems. However, many of the newly §created kernel methods are far from perfect, and§extensions and improvements are always required to §make them even more effective. This book §investigates one important class of the kernel §methods, the least square support vector §machines (LS-SVM), and enhances its §performance extensively. In particular, the LS-SVM §is enhanced in the contexts of four sub-problems §related to solving the pattern classification §problem. That is, model selection, feature §selection, building sparse kernel classifier and §kernel classifier ensemble. The LS-SVM can be §regarded as a representative of many other kernel §methods, and thus many ideas presented in this book §can be easily extended to enhance performance of §those related kernel methods. The results obtained §should be useful to professionals that work on the §theoretical aspects of kernel methods, or anyone §else who may be considering ustilizing kernel §methods for real-world pattern classification §problems.