Selecting Parameters of an Improved Doubly Regularized Support Vector Machine based on Chaotic Particle Swarm Optimization Algorithm

Taking full advantages of the L1-norm support vector machine and the L2-norm support vector machine, a new improved double regularization support vector machine is proposed to analyze the datasets with small samples, high dimensions and high correlations in the parts of the variables. A kind of smooth function is used to approximately overcome the disdifferentiability of the L1-norm and the steepest descent method is used to solve the model. But the parameters of this model bring some trouble about the accuracy of the experiments. By use of the characteristics of chaotic systems, we propose a chaotic particle swarm optimization to select the parameters in the model. Experiments show the improvement gains the better effects.