Fuzzy support vector machine based on density with dual membership

In this paper, an improved fuzzy membership function determination is proposed to train the fuzzy support vector machine (FSVM) for classification which the sample set in reality environment is increasing, and it often contains a lot of noise and outliers. In the improved algorithm, the sample points have the different types of memberships in different regions. The dual membership is introduced to reduce the algorithm complexity and shorten its training time compared with fuzzy support vector machine based on density (DFSVM), at the same time the algorithm well improves the SVM´s accuracy rate.