Research On Theory And Algorithms Of Fast Twin Support Vector Machine

Twin Support Vector Machine(TWSVM)is a new machine learning method based on support vector machine and generalized eigenvalue support vector machine.It is an effective method for rapid classification.For a two-class problem,the TWSVM aims to find a pair of non-parallel hyperplanes for classification.Although TWSVM has a similar expression to SVM,its computational efficiency is approximately 4 times that of SVM.Due to the fast learning ability and good learning performance of TWSVM,it has become a research hotspot in the field of machine learning.However,due to the short development time of the TWSVM,there are still many deficiencies in TWSVM,so it needs further research and improvement.In this paper,the basic contents of SVM and TWSVM are discussed in detail,and then the problems existing in the existing methods are explained by further analysis.In view of these problems,the relevant theories are applied to the in-depth study.The specific research contents are as follows:1.The non singularity method of TWSVM is studied.In order to solve the problem that some research work can't be further carried out in TWSVM,a new nonsingular TWSVM model is proposed by adding the "zero term" method to overcome the singularity of the solution.At the same time,the successive over relaxation method is used to further improve the computing speed of TWSVM.This method can overcome the singularity problem and improve the classification performance.2.The influence of interval distribution on TWSVM is studied.The research shows that interval distribution has an important influence on generalization ability of models.The interval mean and interval variance are introduced in the objective function of standard TWSVM.At the same time,a regularization term based on structural risk minimization is introduced.In order to improve the computational efficiency,the least squares method is used to transform the two programming problems into linear equations,and the Least Squares Large Margin Twin Support Vector Machine are proposed.The algorithm not only has high classification accuracy,but also has fast computation speed.3.Fuzzy least squares twin support vector machine is studied.In order to improve the classification accuracy of TWSVM,the concept of fuzzy is introduced and weighted density function is used to replace the membership function in objective function.At the same time,the Fuzzy Least Squares Large Margin Twin Support Vector Machine(FLS-LMTSVM)is proposed by using the least square method.Theoretical analysis shows that the influence of data set's density on classification results is very important.Therefore,the proposal of FLS-LMTSVM provides an effective way for non-uniform sample problem.