兰军 严广乐

摘要信用评估是商业银行控制和防范信贷风险的关键途径.决策树模型较好的直观解释性使其成为当前个人信用评估中的常用模型，但决策树模型存在容易导致过拟合且预测精度有限的问题.通过在决策树模型算法中引入类随机森林随机有放回的抽样模式，运用机器自动循环迭代寻求最优树的建模思想，建立了自适应最优C5.0决策树个人信用评估模型.该模型具有快速收敛特征变量、较好的泛化能力和高预测精度的特点，在实证分析中对商业银行个人信用评估模型质量提升带来比较明显的改进效果.

关键词数量经济学；个人信用评估；决策树；随机森林；迭代

中图分类号N945文献标识码A

Research and Empirical Analysis on Personal

Credit Evaluation Model Based on Iterative Adaptive

Random Decision Tree Algorithm

Jun Lan，Guangle Yan

（Management School University of Shanghai for Science and Technology， Shanghai200093 China）

AbstractCredit evaluation is a crucial approach to control and reduce credit risk. The direct expression feature of decision tree model has made it a commonly used model in individual credit evaluation. However， this model has the disadvantages of over fitting and limited accuracy. By utilizing random forest sampling with replacement in decision tree model， with autoiteration for optimal result， an autofitting optimal C50 decision tree for individual credit evaluation model is established. The model has features such as rapidly converging character variables， good generalization ability and high predicting accuracy， and is shown to have phenomenal improvement on quality of individual credit evaluation in real practical analysis.

Key wordsmathematical economics； personal credit assessment； decision tree； random forest； iteration

1引言

近年來，随着中国经济的飞速发展，个人的信用状况越来越受到重视，贷款购房、贷款购车、申领信用卡等各种个人消费贷款都需要了解个人的信用信息.个人信用制度是国家信用体系的一部分，科学合理有效的个人信用评估体系是个人信用体系的核心问题之一.对个人信用状况进行科学评估，是发展个人消费信贷的基础，也是中国市场经济发展的有益补充和完善.中国已经加入世界贸易组织（WTO）多年，逐步开放的信用市场，对建立个人信用评估体系提出了严峻挑战.当前中国个人信用制度不完善，个人信用信息欠完备，个人信用难以评估，已经比较严重地影响了经济增长.如何建立一个统一、科学的个人……