摘要
本文以德国某商业银行的真实客户信用数据为样本,将决策树方法应用于个人信用指标的筛选过程中,并与BP神经网络模型相结合构建成一个两阶段组合模型。研究表明,基于决策树—神经网络构建的个人信用评估组合模型对于测试样本的分类预测精度高于单一BP神经网络模型的分类预测精度。组合模型对于测试样本的总正确率平均值为75.45%,高出单一BP神经网络模型的总正确率近3个百分点。基于信息熵增益率分类原理的最优决策树挑选指标方法能合理去除非重要属性指标的干扰,使真正有效的属性指标输入神经网络主模型,提高模型分类预测的精准度。
The real customers＇ credit data of a German commercial bank as samples, this paper applies decision-tree method in the selection process of personal credit indicators. The decision tree method is combined with the BP （Back Propagation） neu- ral network model to form a two-phase composite model. The study of the paper shows that, for the test samples, the classifica- tion prediction of the combined model, used for personal credit rating and based on decision tree-neural network, is more accu- rate than that of the single BP neural network model and the average of total correct rates is 75.45%, nearly 3 percentage higher than the single BP neural network model. The indicator selection of optimal decision tree, based on the information-entropy- gain classification principle, can rationally eliminate the interference of unimportant attribute indicators, indroduce the truly ef- fective attribute indicators to enter the main model of the neural network and improve the accuracy of the classification prediction of the model.
出处
《金融论坛》
CSSCI
北大核心
2013年第2期57-61,67,共6页
Finance Forum
基金
国家社会科学基金重大项目<加快社会信用体系建设研究>(12&ZD053)
国家社会科学基金重点项目<社会诚信制度建设和维护市场经济秩序问题研究>(11AZD026)
国家自然科学基金面上项目<社会信用制度建设关键技术
建设标准与实现机制研究>(71073047)的阶段性成果
关键词
个人信用评估
决策树
神经网络
组合模型
personal credit rating
decision tree
neural network
combined model
