Research on SVM Multi-Classification Based on Particle Swarm Algorithm

Abstract

In order to solve deviation and unbalance problem generally in traditional multi-class classification, on the basis of mutual communication entropy theory and classification principle of support vector data description (SVDD), this paper designs a kind of improved locality SVDD multi-class classification algorithm, namely, EL-SVDD algorithm. This algorithm firstly takes local sample information as the carrier to calculate the mutual communication entropy parameter value; second in the multidimensional sphere, it classifies mutual communication entropy parameter values to place test sample data information; finally, analyze the test sample size and mutual communication entropy parameter values comprehensively, to reinterpret the C value in SVDD algorithm. Experiments show that the EL-SVDD algorithm is not only feasible, but also can effectively and steadily improve the multi-class analysis accuracy.