Abstract: The conception of fuzzy membership is introduced into least square support vector machines(LSSVMs), which overcomes the disadvantage that LSSVMs are so sensitive to outliers in training samples. And then fuzzy least square support vector machines (FLSSVMs) are proposed based on support vector domain description (SVDD). Data samples in the feature space are described and the smallest enclosing hypersphere is obtained. The fuzzy membership value to each sample point is determined according to the distance of each sample from the center of the hypersphere, which can reduce the effect of outliers. Numerical results show that the predictive precision of the proposed method is higher than that of SVMs and LSSVMs without decreasing the training speed.

Key words: least square support vector machines, fuzzy membership, data domain description