Training fuzzy support vector machines by samples in zonal regions

Fuzzy support vector machines based on zonal regions are constructed by using potential points that likely being the support vectors. Firstly, two nearest samples, including one positive sample and one negative sample in the training set, are selected to construct rough classification hyperplane. Secondly, all training samples are mapped to the zonal regions by their distances to the rough classification hyperplane, and the suitable threshold λ is used to select the samples being likely support vectors, which are composed of the zonal regions. Finally, fuzzy support vector machines are constructed on the zonal regions. The experiment results on machine learning benchmark testing sets show that the proposed learning machines not only reduce the number of training samples and training time, but also improve generalization ability of the learning machines.