An improved fuzzy twin support vector machine based on support vector

Abstract

Twin support vector machine (TWSVM) has faster speed than traditional support vector machine (SVM) for classification problem. However, it ignores the effect of noise samples on the optimal hyperplanes with respect to the classification task. Membership functions of traditional fuzzy twin support vector machine (FTSVM) are mostly designed based on the distance between the samples and the class centers, which decreases the effect of support vectors. In this paper, the new membership function divides the samples into three parts: support vectors, non-support vectors and outliers. An improved fuzzy twin support vector machine (IFTSVM) is proposed based on support vectors. IFTSVM is an improved TWSVM with better classification performance. Experimental results show that IFTSVM is more effective and feasible in classification.