客户信用评估模型  

Customer credit scoring model

作　　者：赵亚鹏[1] 丁以中[1] 

机构地区：[1]上海海事大学交通运输学院,上海200135

出　　处：《上海海事大学学报》2008年第2期58-61,共4页Journal of Shanghai Maritime University

基　　金：上海市重点学科建设项目(T0602);上海市教育委员会重点项目(06ZZ43)

摘　　要：为克服由因于客户信用评估的非线性和不确定性,且样本数据积累少、偏差大和真实数据获得难度较大而产生的困难,提出1种基于改进的遗传神经网络客户信用评估模型.将该模型应用于客户信用评估研究及试验均表明,基于改进的遗传神经网络客户信用评估模型在模型分类准确率和分类准确率的标准偏差两方面均明显优于Logit,K-NN和BP神经网络客户信用评估模型,并有效地解决样本量少和偏差大的问题,显著提高信用评估模型的推广能力,具有良好的稳健性和精度.In order to overcome the difficulties of customer credit scoring such as nonlinearity, uncertainty, few sample data accumulation, large deviation and the problem that real data is hard to obtain, an improved genetic neural network model is proposed to evaluate customer credit risk. It is showed by research and experiment that the improved genetic neural network model is much better than Logit, K-NN and BP network models on classification accuracy and standard deviation of classification accuracy. The problems of few samples and large deviation of customer credit risk assessment are availably solved. The generalization of the model with good stability and accuracy is significantly improved.

关 键 词：信用评估 遗传算法 神经网络 偏差 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]