Construction and application of PSO-SVM model for personal credit scoring

Abstract

The parameters of support vector machine (SVM) are crucial to the model's classification performance. Aiming at the randomicity of selecting the parameters in SVM, this paper constructed a PSO-SVM model by using particle swarm optimization (PSO) to search the parameters of SVM. The model was used for personal credit scoring in commercial banks and particles' fitness function was used to control the type II error which costs huger loss to commercial banks. Compared with BP NN, the application results indicate that PSO-SVM gets higher classification accuracy with lower type II error rate and the model shows stronger robustness, which presents more applicable for commercial banks to control personal credit risks. © Springer-Verlag Berlin Heidelberg 2007.