A new fuzzy twin support vector machine for pattern classification

Fuzzy SVM is often used to solve the problem that patterns belonging to one class often play more significant roles in classification. In order to improve the efficiency and performance of fuzzy SVM, this paper proposes a new fuzzy twin support vector machine (NFTSVM) for binary classification, in which fuzzy neural networks and twin support vector machine (TWSVM) are incorporated. By design, the influence of the samples with high uncertainty can be mitigated by employing fuzzy membership to weigh the margin of each training sample, which improves the generalization ability. In addition, we show that the existing TWSVM and twin bounded support vector machines (TBSVM) are special cases of the proposed NFTSVM when the parameters of NFTSVM are appropriately selected. Moreover, the successive overrelaxation (SOR) technique is adopted to solve the quadratic programming problems (QPPs) in the proposed NFTSVM algorithm to speed up the training procedure. Experimental results obtained on several artificial and real-world datasets validate the feasibility and effectiveness of the proposed method.