遗传繁衍样本策略及神经网络信用评价研究  

A Strategy of Generating Samples Based on Genetic Algorithm and Its Application on Credit Score by Neural Networks

作　　者：吴德胜[1] 梁梁[1] 

机构地区：[1]中国科学技术大学商学院,安徽合肥230026

出　　处：《管理科学》2004年第1期59-65,共7页Journal of Management Science

基　　金：国家863项目基金资助项目(2002AA41361);国家自然科学基金资助项目(70371023)

摘　　要：提出遗传算法繁衍样本策略,克服了小样本建立模型的缺陷,建立适合于我国商业企业的信用评分指标体系,依据该指标体系和遗传优化样本建立了基于反向传播回归神经网络的信用评估模型,利用样本公司实际指标数据对该模型的评分效果进行了比较研究。A strategy of generating samples based on genetic algorithms is proposed, thus the deficiency of small-sample-modeling is handled. Credit-score models suitable for Chinese corporations are built. On this basis together with the genetic optimized samples, credit evaluation model is established by means of BP(Backpropagation) neural networks. Finally, acomparative study of the credit-score models' scoring effects is carried out using actual index data.

关 键 词：信用评分 神经网络 遗传算法 反传算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]