Improvements On Twin Support Vector Machines And Its Application

Twin Support Vector Machines were first proposed by Jayadeva and others in 2007,whose basic idea is to construct a hyperplane for each of the two classes of training points,let each hyperplane with one kind of training points as close as possible,and away from another class of training points.The new training sample point is close to which hyperplane,it belongs to this classification.Compared to the classical support vector machine,TWSVM converted a large quadratic programming problem into two small quadratic programming problem,making the training time greatly reduced,and the computing efficiency is about four times that of SVM.In view of its excellent learning performance,TWSVM has become a research hotspot in the field of machine learning and data mining.After in-depth study and research on TWSVM,it is found that TWSVM can not reduce the influence of noise on the optimal hyperplane,and TWSVM is originally designed for two classification problems,and lacks the ability to deal with multi class classification problems.In view of the above problems,this paper has done the following work:The principle of two arithmetic of SVM and TWSVM is studied,and its main reasoning process is introduced.By contrast,the good performance of TWSVM is proved.In order to reduce the influence of noise or wild points on TWSVM optimal hyperplane,three improved twin support vector machines are proposed.Experiments on artificial data and multiple sets of UCI data demonstrated that these three improved methods can effectively improve the classification accuracy of TWSVM.In this paper,a hybrid binary tree twin support vector machine is proposed by constructing a reasonable binary tree structure and combining the improved dual support vector machine,in order to extend the excellent performance of TWSVM to the multi class classification problem.Finally,the good classification performance is proved by numerical experiments.