Abstract:

We improve the membership setting method for the fuzzy support vector machine (FSVM). Given the distribution uncertainty of training sets, we replace the classical Euclidean distance with grey incidence degree, and define the mean grey absolute incidence degree of samples. To overcome the disadvantages in traditional methods and ensure a greater contribution of support vectors to the classification results, we present a new approach for distinguishing noises based on the dual reference points of the same center and the different center. The relationship between the parameters in solving the FSVM and the corresponding support vectors is analyzed, and the steps of setting  the fuzzy membership are given. Experimental results demonstrate the effectiveness and practicability of this approach.

Key words: fuzzy support vector machine；grey incidence degree；membership；dual reference points