Parameter selection method for SVM with PSO

In purpose of automatically tuning multiple parameters for Support vector machine (SVM), a parameter selection method is proposed for SVM based on Particle swarm optimal (PSO) algorithm. In our method, each particle indicates a choice of multiple parameters, the population is a collection of particles, and the new method only requires the evaluation of an objective function to guide its search without additional derivatives or auxiliary knowledge required. The number ratio of support vectors to training samples is used to estimate the generalization performance. The new method is tested on different sizes of benchmark datasets with binary class problem. Simulation results demonstrate the effectiveness of the proposed method.