Abstract: In order to improve the problem of inefficient parameter selection of the GDS-SVC model and DIPSO-SVC model, and utilize the generalization ability and robustness of support vector classification (SVC), the reduction factor of location updating was introduced based on the dynamic improvement Particle Swarm Optimization (DIPSO), and then the DKIPSO-SVC of parameters selecting in SVC was established based on DKIPSO. The method was applied to credit scoring of commercial banks. The simulation results demonstrate that the robustness of the DKIPSO-SVC model is better than DIPSO-SVC. But beyond that, the accuracy of DKIPSO-SVC model achieves 96.6049%, higher than that of DIPSO-SVC and GDS-SVC model, which is 93.8272% and 92.5926%. More importantly, the type II error rate was reduced significantly from 8.5526% to1.9737%, about 76.9228% lower than current model.

Key words: credit scoring, support vector machine, particle swarm optimization, DKIPSO-SVC model