基于支持向量机的信用评估模型及风险评价

运用基于支持向量机理来建立一个新的个人信用评估预测模型,以期取得更好的预测分类能力.并对SVM分类结果与三层全连接BPN分类结果进行了比较.结果表明,在判别潜在的贷款申请者中支持向量的判别结果比神经网络的要好.为了减小训练集偏差及为了验证两种方法的鲁棒性,基于两种策略(平衡样本与非平衡样本)交叉验证来进一步评价SVM分类准确性,并对两种方法基于两种策略的误分类作了风险代价分析.

Credit scoring models and credit-risk evaluation based on support vector machines

A new credit-scoring was developed to provide a new better judgment method,based on support vector machine(SVM) models that accurately classify consumer loan applications.This study also compared the performance of SVM and threelayer fully connected back-propagation neural networks(BPN) in credit scoring. The SVM models consistently performed better than the BPN models in identify potential problem loans. To alleviate the problem of bias in the training set and to examine the robustness of SVM classifiers in identifying problem loans,we cross-validate our results through two different strategies(no-balance sample data set and balance sample data set).In addition,we estimated risk cost of credit scoring error for two models.