Fuzzy output support vector machines for classification

Support vector machines just use the sign of decision value to get the decision class but don't take its value into consideration. Compared with the support vector machines, the proposed machine not only gives the decision class, but also the membership to each class using the decision value. For SVMs are essentially a 2-class classifier, we first construct the fuzzy output SVMs for 2-class, then extend it to multi-class case. In multi-class case, the feature space is divided into three parts: absolutely classified region, unclassified region and positive margin region because of different accuracy in them. In different regions, the range of the value of membership is different. Through the membership, we can get the location information of the data, which can tell us the confidence of the decision. So this will be helpful for further decision and analysis. The experiments show that the performance of fuzzy output SVMs is almost the same as the one-to-one approach, but when the membership to two classes is comparative and less than 0.8, the second maximal membership can sometimes correspond to the real class.