«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

32

期数:

2022年06期

页码:

45-51

栏目:

大数据分析与挖掘

出版日期:

2022-06-10

文章信息/Info

Title:

Personal Credit Evaluation Method Based on SMOTE+ENN

文章编号:

1673-629X(2022)06-0045-07

作者:

吕 颖; 邢进生

山西师范大学 数学与计算机科学学院,山西 临汾 041004

Author(s):

LYU Ying; XING Jin-sheng

School of Mathematics and Computer Science,Shanxi Normal University,Linfen 041004,China

关键词:

信用评估; 数据不平衡; 数据预处理; 网格搜索; 集成学习

Keywords:

credit evaluation; data imbalance; data preprocessing; grid search; ensemble learning

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2022. 06. 008

摘要:

个人信用评估作为商业银行判定借贷风险的直接依据,在金融领域显得尤为重要。 针对传统个人信用评估模型存在数据不平衡、模型结构单一、易受主观因素干扰等问题,提出一种基于 SMOTE + ENN( synthetic minority oversamplingtechnique+edited nearest neighbours)算法与集成学习的个人信用评估方法。 首先,该方法在数据预处理的基础上,采用SMOTE+ENN 算法对样本数据进行数据平衡分布处理,增强了分类算法性能;然后,基于网格搜索优化算法,搜寻适用于多种分类器的最优超参数,进而构造出相应的最优单一评估模型,达到了提高个人信用评估精确度的目的;最后,利用相关的集成学习策略将表现最优的三种分类器结果集成,构造出信用评估的最优预测模型,从而实现更为准确的个人信用评估。 实验结果表明,在现有公开数据集 Give Me Some Credit 上,与传统数据不平衡处理方法相比,该方法的预测准确率高达 97% ,精确度提升约 2% ,验证了算法改进的有效性。

Abstract:

Personal credit evaluation as a direct basis for commercial banks to judge loan risk is particularly important in the financialfield. Aiming at the problems of the traditional personal credit evaluation model,such as data imbalance,single model structure and beingeasily interfered by subjective factors,a personal credit evaluation method based on SMOTE + ENN algorithm and ensemble learning isproposed. First of all,SMOTE+ENN algorithm is used to balance and distribute the sample data on the basis of data preprocessing,whichenhances the performance of the classification algorithm. Then, based on the grid search optimization algorithm, the optimal superparameters suitable for a variety of classifiers are searched, and the corresponding optimal single evaluation model is constructed toachieve the purpose of improving the accuracy of personal creditevaluation. Finally,the results of the three classifiers with the best performance are integrated with the related ensemble learning strategy to construct the optimal prediction model of credit evaluation,so as toachieve a more accurate personal credit evaluation. Experiment shows that on the existing public dataset Give Me Some Credit,comparedwith the traditional data imbalance processing method,the proposed method is as high as 97% in prediction accuracy,and the accuracy isimproved by about 2% ,which verifies the effectiveness of the improved algorithm.

常用功能

更新日期/Last Update: 2022-06-10