Personal Credit Scoring Model Based on SVM Optimized by GA

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 presents a method to optimize the parameters of SVM by using genetic algorithm (GA). Using GA's global search to optimize the parameters of SVM and using the chromosome's fitness function to control the type II error rate in personal credit scoring which costs great loss to commercial banks, compared with BP neural network, the application results indicate that SVM model optimized by GA gets higher classification accuracy and the type II error rate is limited efficiently. The SVM model optimized by GA also shows stronger robustness which presents more applicable for commercial banks to control the consumer credit risks.