Improvement On Fuzzy Twin Support Vector Machine And Its Solving Method

Support vector machine developed based on statistical learning theory, which is built on the basis of the principle of minimal VC dimension theory and structural risk. Researchers propose improvement theroy on SVM through extensive research such as fuzzy support vector machine, twin support vector machines and fuzzy twin support vector machine. The methods not only can improve the accuracy, but also can reduce running time. To improve the performance of support vector machine, this paper propose fuzzy twin bounded support vector machine and do some research on solving the Lagrange multipliers of fuzzy twin bounded support vector machine. Specific studies are the following:1. Improvements on fuzzy twin support vector machine algorithmIn-depth research has been made on fuzzy support vector machine, twin support vector machines and fuzzy twin support vector machines. We proposed fuzzy twin bounded support vector machine(FTBSVM) through improve fuzzy twin support vector machine. The method not only considered the empirical riskand structural risk, but also considered the importance of data and the influence of the noise sample points on the surface. The classification correct rate is superior than FTSVM.2. The solving method of Lagrange multipliers on fuzzy twin support vector machine and fuzzy twin bounded support vector machine It usually need to solve a quadratic programming problem to botain the Lagrange multipliers for fuzzy twin support vector machine and fuzzy twin bounded support vector machine.. But solving the problem needs a long time and higher cost. So in order to improve the performance of fuzzy twin support vector machine and fuzzy twin bounded support vector machine we use successive overrelaxation which is converged to obtain the Lagrange multipliers.3. Some experiments have been made to study the performance of fuzzy twin support vector machine and fuzzy twin bounded support vector machine we choose the standard dataset from UCI and use tenfold cross-validation method to study the performance of fuzzy twin support vector machine and fuzzy twin bounded support vector machine. Finally the experiments using traditional method also be made to compare with successive overrelaxation(SOR) method.