Optimization of combined kernel function for SVM by Particle Swarm Optimization

To choose an appropriate kernel function is one major task for SVM. Different kernel functions will produce different SVMs and may result in different performances. Combined kernel function shows more stable and higher performance than single kernel function, so there is a need to optimize the combined kernel function to enhance the generalization capability of SVM. This paper proposes to optimize the combined kernel function by particle swarm optimization (PSO) based on large margin learning theory of SVM. The comparison of the performance between GA and PSO algorithm on this optimization problem is provided. The simulation results show that the PSO is another feasible solution for optimization of combined kernel function, which normally leads to SVM with better generalization capability and stability.