Abstract:  A new fuzzy support vector machine (SVM) model with dual membership values, called DFSVM in short, is developed for the credit evaluation of smallandmediumsized enterprises. In this model, each sample belongs to two credit classes according to its dual membership values. The optimal input indicator portfolios are determined by using the attribute reduction method in rough set theory. With banks credit risk aversions taken into account, samples in the two classes are handled asymmetrically in the training process. Empirical results show that the discrimination accuracy of the proposed DFSVM model is superior to the traditional discrimination models. Furthermore, an adjusted model is proposed, test shows that the adjusted DFSVM model can further reduce the banks credit risk.

Key words: rough set, support vector machine (SVM), credit evaluation, dual membership values