Performance Comparison For Several Classes Of SVMs

This thesis mainly deals with the performance comparisons of several representative linear support vector classification machines which are popular nowadays.The purpose of it is to provide specific and targeted choices for practical application.By introducing kernel function and nuclear techniques,similar approaches can be used to compare the performances of nonlinear support vector machines.Therefore,there will be no more discussion about it in this thesis.To start with,the performance of support vector machines will be compared based on the binary classification problems.Next comes the comparison of them based on the multi-class classification problems.The whole thesis is divided into four chapters which are as follows.Chapter 1 mainly introduces the preliminary knowledge,including some basic concepts as well as the classical hard-margin support vector machine and soft-margin support vector machine.Chapter 2 aims at the binary classification problems.From the aspects of sparse,learning rate,accuracy rate,empirical risk minimization,the structural risk minimization and matrix singularity,the author of the thesis compares the performances of a variety of representative support vector machines,namely soft-margin SVM,least squares SVM,Smooth SVM,twin SVM,twin bounded SVM,Projection TSVM,Lagrangian TSVM,Weighted TSVM,Weighted LS-PTSVM,Bi-density TSVM,Laplacian TSVM.The multi-class classification problems are chiefly discussed in the third chapter.The first part is the illustration of the six strategies applied in solving the multi-class classification problems of the support vector machines and K(K?3)-class TSVM.Next are the performance comparisons of Twin K(K?3)-class support vector machine,least squares twin KSVM,one-versus-all least squares twin KSVM,one-versus-one least squares twin KSVM,weighted loss twin KSVM and weighted loss of multi-class twin support vector machines based on information granularity.The fourth chapter focuses on the brief conclusion and expectation of Twin Support Vector Machine.