A New Fuzzy Membership Assignment Approach for Fuzzy Svm Based on Adaptive Pso in Classification Problems


Omid Naghash Almasi - Young Researchers and Elite Club, Mashhad Branch, Islamic Azad University, Mashhad, Iran Hamed Sadeghi Gooqeri - Department of Electrical Engineering, Bandar Abbas Branch, Islamic Azad University, Bandar Behnam Soleimanian Asl - Department of Electrical Engineering, Gonabad Branch, Islamic Azad University, Mashhad, Iran Wan Mei Tang - Department of Computer and Information Science, Chongqing Normal University, Chongqing


Noises will confuse Support Vector Machine (SVM) in the training phase. To overcome this problem, SVM was extended to Fuzzy SVM (FSVM) by incorporating an appropriate fuzzy membership to each data point. Thus, how to choose a proper fuzzy membership is of paramount importance in FSVM. In this paper, Adaptive Particle Swarm Optimization (APSO) method minimizes the generalization error by changing the attributes values of positive and negative class centers to make them free of attribute-noise. As the APSO converged, the fuzzy memberships are assigned for each training data points based on their distance to the corresponding purified class centers with the same class-label. To demonstrate the effectiveness of the proposed FSVM, its performance on artificial and real-world data sets is compared with three FSVM algorithms in the literature.