Abstract: An Affinity Class-Hyperparallel Fuzzy Support Vector Machine was proposed to get better classification result. This method not only takes the advantage of the affinity, but also abandons the estimation that the samples obey spherical-shape distribution. Instead of the cluster center, a hyperplane within the class is used to find a hyperparallel with the minimum distance while containing the maximum samples by the way of quadratic programming. The membership is achieved through a new S-function based on the distance of the hyperparallcl which reflects the affinity of the samples. The simulation on UCI shows that the ACHFSVM is more robust and has better classification accuracy.

Key words: Fuzzy support vector machine, Affinity, Fuzzy membership, Classification