Fast and robust learning through fuzzy linear proximal support vector machines

Traditional support vector machines (SVMs) assign data points to one of two classes, represented in the pattern space by two disjoint half-spaces. In this paper, we propose a fuzzy extension to proximal SVMs, where a fuzzy membership is assigned to each pattern, and points are classi1ed by assigning them to the nearest of two parallel planes that are kept as distant from each other as possible. The algorithm is simple and fast, and can be used to obtain an improved classi1cation when one has an estimate of the fuzziness of samples in either class.