For the radial basis function (RBF) kernel based support vector machines (SVM), particle swarm optimization (PSO) is employed to carry out the model optimization. The value space of the parameter is presented on the analysis of the mean shortest distance and mean furthest distance among samples of the training set thus the search region is reduced, logarithmic scale is employed to further improve the search efficiency of PSO. Extensive experiments on comparison with genetic algorithm and grid based approaches indicate that the proposed approach converges faster and produces better hyper-parameters.