A Novel Classification Algorithm Based on Fuzzy Kernel Multiple Hyperspheres

In this paper a novel classification algorithm based on fuzzy kernel multiple hyperspheres is presented. In the training process all training samples of each class are covered by the constructed multiple hyperspheres. Each hypersphere encompasses as many samples with the same class as possible via the greedy method. A fuzzy membership function is defined to label the testing samples in the classification process. Moreover, the kernel function is used instead of Euclidean inner product. Finally, Experiments on three artificial datasets and five real datasets show that our approach is valid and has encouraging pattern classification performance.