Fuzzy Support Vector Machines Based on Density Clustering

The improved fuzzy support vector machines (IFSVMs) are proposed in this paper. The proposed learning machines select the sparse data in each class to training FSVMs. First the proposed methods select the relative sparse training data by using the suitable parameters, the radii and the size of the area. Second, as the representation of the entire training data, the selected sparse training data are used to train the IFSVMs. Third, the integration of two kinds FSVMs is used to verify the performance of the proposed learning machines. The simulation results on the benchmark datasets of machine learning databases show that the IFSVMs not only downsize the training set but also reduce the running time and hardly influence on the generalization ability of learning machines.