基于Focal Loss修正交叉熵损失函数的信用风险评价模型及实证

doi:10.16381/j.cnki.issn1003-207x.2020.2188

中国管理科学 ›› 2022, Vol. 30 ›› Issue (5): 65-75.doi: 10.16381/j.cnki.issn1003-207x.2020.2188

• 论文 • 上一篇

基于Focal Loss修正交叉熵损失函数的信用风险评价模型及实证

杨莲^1,2, 石宝峰^1,2

1.西北农林科技大学经济管理学院，陕西杨凌712100；2.西北农林科技大学信用大数据应用研究中心，陕西杨凌712100

收稿日期:2020-11-20 修回日期:2020-12-15 发布日期:2022-06-01
通讯作者: 石宝峰(1984-)，男(汉族)，山西长治人，西北农林科技大学经济管理学院,教授、信用大数据应用研究中心主任，博士生导师，研究方向：风险管理、普惠金融，Email：shibaofeng@nwsuaf.edu.cn. E-mail:shibaofeng@nwsuaf.edu.cn
基金资助:
国家自然科学基金资助面上项目(72173096，71873103)；国家自然科学基金资助重点项目(71731003)；中央农办、农业农村部乡村振兴专家咨询委员会软科学项目(202122)；陕西省社会科学基金资助项目(2018D51)；陕西省创新人才推进计划青年科技新星项目(2019KJXX-070)；中和农信“星空计划”项目(K4030218167)；西北农林科技大学仲英青年学者项目(2021-04)

Credit Risk Evaluation Model and Empirical Research Based on Focal Loss Modified Cross-Entropy Loss Function

YANG Lian^1,2, SHI Bao-feng^1,2

1. College of Economics and Management, Northwest A&F University, Yangling 712100, China;2. Research Center on Credit and Big Data Analytics, Northwest A&F University, Yangling 712100, China

Received:2020-11-20 Revised:2020-12-15 Published:2022-06-01
Contact: 石宝峰 E-mail:shibaofeng@nwsuaf.edu.cn

摘要/Abstract

摘要： 针对信用评价违约、非违约样本比例失衡，容易出现评价模型对非违约样本识别过度，对违约样本、尤其是违约样本中困难样本识别不足的问题，将图像识别中得以广泛应用的焦点损失Focal Loss函数引入信用评价，构建Focal Loss修正交叉熵损失函数的信用风险评价模型，并用三个数据集验证了模型的有效性。创新与特色：一是在信用评价交叉熵损失函数中引入聚焦参数γ构造调节因子(1－y′)γ，通过增大困难样本在目标损失中的权重，构建ADASYN-BPNN-FocalLoss信用风险评价模型，保证信用评价模型对不均衡数据中违约样本的识别力，弥补了现有深度学习信用评价模型无法有效识别不均衡数据中困难样本的不足。二是通过测算违约样本的K近邻非违约样本占比ri，求解需新合成的样本数gi，进而利用SMOTE算法合成新的违约样本，既保证了新生成的违约样本si能够反映原信用评价数据的基本特征，也改变了现有违约、非违约样本不均衡致使评价模型判别能力偏低的现状。三是利用本文所建模型与ADASYN-BPNN-CrossEntropy、决策树、K最近邻、随机森林等5种模型，对中国1298个农户贷款数据和UCI公开的德国、澳大利亚信贷数据集进行分析，实证表明本文所建模型AUC、Type2-error等指标均优于现有模型。该方法可有效提升模型对困难样本的识别能力，改善违约预测性能。

关键词: 信用评价；Focal Loss；BP神经网络；自适应综合过采样

Abstract: Credit evaluation model plays an important role in helping financial institutions to identify default risk. However, due to the imbalance of the proportion of default and non-default samples, there are the phenomena of over-recognition for non-default samples and under-recognition for default samples. Some of default samples, named hard samples, are difficult to be identified. Therefore, the key to improve the prediction performance of the model is to improve its ability to recognize the hard samples. In practice, the existing deep learning credit evaluation model, which takes the Cross Entropy as the loss function, considers that there is no difference between the contribution of the hard samples and the simple samples to the target loss. It affects the effective identification of hard samples by the model.

Key words: credit evaluation; focal loss; BP neural network; adaptive comprehensive oversampling

中图分类号:

杨莲, 石宝峰. 基于Focal Loss修正交叉熵损失函数的信用风险评价模型及实证[J]. 中国管理科学, 2022, 30(5): 65-75.

YANG Lian, SHI Bao-feng. Credit Risk Evaluation Model and Empirical Research Based on Focal Loss Modified Cross-Entropy Loss Function[J]. Chinese Journal of Management Science, 2022, 30(5): 65-75.

参考文献

［1］ Imai K, Gaiha R, Thapa G, et al. Microfinance and poverty-a macro perspective［J］. World Development, 2012, 40(8): 1675-1689.
［2］程砚秋, 徐占东. 基于泰尔指数修正的ELECTRE Ⅲ小企业信用评价模型［J］. 中国管理科学, 2019, 27(10): 22-33.Cheng Yanqiu, Xu Zhandong. Credit risk evaluation of small enterprises based on revised ELECTRE III by Theil index［J］. Chinese Journal of Management Science, 2019, 27(10): 22-33.
［3］ Lin T, Goyal P, Girshick R, et al. Focal loss for dense object detection［J］. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 99: 2999-3007.
［4］李哲, 迟国泰. 基于最大指标区分度与最优相对隶属度的上市公司信用风险研究［J］.中国管理科学, 2021, 29(4): 1-15.Li Zhe, Chi Guotai. Research on the listed companies’ credit risk based on maximum discrimination and optimal relative membership degree［J］. Chinese Journal of Management Science, 2021, 29(4): 1-15.
［5］迟国泰, 李鸿禧. 基于逐步判别分析的小企业债信评级模型及实证［J］. 管理工程学报, 2019, 33(4): 205-215.Chi Guotai, Li Hongxi. Debt rating model of small businesses and empirical analysis based on stepwise discriminant［J］. Journal of Industrial Engineering and Engineering Management, 2019, 33(4): 205-215.
［6］ Guo Yanhong, Zhou Wenjun, Luo Chunyu, et al. Instance-based credit risk assessment for investment decisions in P2P lending［J］. European Journal of Operational Research, 2016, 249(2): 417-426.
［7］ Chen Ning, Ribeiro B, Chen An. Financial credit risk assessment: A recent review［J］. Artificial Intelligence Review, 2016, 45: 1-23.
［8］牟刚, 袁先智. 大数据架构下企业内部信用评级的实证研究［J］. 系统工程学报, 2016, 31(6): 808-815, 849.Mu Gang, Yuan Xianzhi. Empirical study for enterprise internal credit rating under big data framework［J］. Journal of Systems Engineering, 2016, 31(6): 808-815, 849.
［9］吕德宏, 朱莹. 农户小额信贷风险影响因素层次差异性研究［J］. 管理评论, 2017, 29(1): 33-41.Lv Dehong, Zhu Ying. Research on the factors and hierarchy difference of farmer household microfinance risk［J］. Management Review, 2017, 29(1): 33-41.
［10］ Angilella S, Mazzù S. The financing of innovative SMEs: A multicriteria credit rating model［J］. European Journal of Operational Research, 2015, 244(2): 540-554.
［11］衣柏衡, 朱建军, 李杰. 基于改进SMOTE的小额贷款公司客户信用风险非均衡SVM分类［J］. 中国管理科学, 2016, 24(3): 24-30.Yi Baiheng, Zhu Jianjun, Li Jie. Imbalanced data classification on micro-credit company customer credit risk assessment using improved SMOTE support vector machine［J］. Chinese Journal of Management Science, 2016, 24(3): 24-30.
［12］董路安, 叶鑫. 基于改进教学式方法的可解释信用风险评价模型构建［J］.中国管理科学, 2020, 28(9):45-53.Dong Luan, Ye Xin. Interpretable credit risk assessment modeling based on improved pedagogical method［J］. Chinese Journal of Management Science, 2020, 28(9):45-53.
［13］石宝峰,刘锋,王建军,等. 基于 PROMETHEE-II 的商户小额贷款信用评级模型及实证［J］. 运筹与管理, 2017, 26(9): 137 -147.Shi Baofeng, Liu Feng, Wang Jianjun, et al. A credit rating model of microfinance loans for small private business based on PROMETHEE-II and its empirical study［J］. Operations Research and Management Science, 2017, 26(9): 137-147.
［14］ Xia Yufei, Liu Chuanzhe, Li Yuying, et al. A boosted decision tree approach using bayesian hyper-parameter optimization for credit scoring［J］. Expert Systems with Applications, 2017, 78: 225-241.
［15］ Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks［J］. Science, 2006, 313(5786): 504-507.
［16］肖斌卿, 杨旸, 李心丹, 等. 基于模糊神经网络的小微企业信用评级研究［J］. 管理科学学报, 2016, 19(11): 114-126.Xiao Binqing, Yang yang, Li Xindan, et al. Research on the credit rating of small and micro enterprises based on fuzzy neural network［J］. Journal of Management Sciences in China, 2016, 19(11): 114-126.
［17］何珊, 刘振东, 马小林. 信用评分模型比较综述—基于传统方法与数据挖掘的对比［J］. 征信, 2019, 37(2): 57-61.He Shan, Liu Zhendong, Ma Xiaolin. A comparative review of credit scoring models —based on the comparison between traditional methods and data mining［J］. Credit Reference, 2019, 37(2): 57-61.
［18］ Rumelhart D E, Hinton G E, Williams R J. Learning representations by back-propagating errors［J］. Nature, 1986, 323(6088): 533-536.
［19］杨胜刚, 朱琦, 成程. 个人信用评估组合模型的构建—基于决策树-神经网络的研究［J］. 金融论坛, 2013, 18(2): 57-61, 67.Yang Shenggang, Zhu Qi, Cheng Cheng. The building of the combined model for personal credit rating - a study based on the decision tree-neural network［J］. Finance Forum, 2013, 18(2): 57-61, 67.
［20］ Standard & Poor’s Ratings Services. S&P’s study of China’s top corporates highlights their significant financial risks ［R］. Standard & Poor’s Ratings Services, September 13, 2012: 175-199.
［21］ Fitch Ratings. Fitch ratings global corporate finance 2012 transition and default study ［R］. Credit Market Research-Fitch Ratings, March 2013: 2-27.
［22］ Moody’s Investors Service. Rating symbols and definitions ［R］. Moody’s Investors Services, 2016: 1-48.
［23］中国农业银行. 中国农业银行信贷资产风险分类管理办法［R］. 中国农业银行, 2011.Agricultural bank of China. Agricultural bank of China credit asset risk classification management measures ［R］. Agricultural bank of China, 2011.
［24］中和农信项目管理有限公司. 中和农信农户信用评价打分表［R］. 中和农信项目管理有限公司, 2017.Chongho bridge management limited. Chongho bridge farmers’ credit evaluation scoring table ［R］. Chongho bridge management limited, 2017.
［25］石宝峰, 王静, 迟国泰. 普惠金融、银行信贷与商户小额贷款融资—基于风险等级匹配视角［J］. 中国管理科学, 2017, 25(9): 28-36.Shi Baofeng, Wang Jing, Chi Guotai. The inclusive finance, bank loans and financing of small private business microfinance loan［J］. Chinese Journal of Management Science, 2017, 25(9): 28-36.
［26］石宝峰, 王静. 基于ELECTRE III的农户小额贷款信用评级模型［J］. 系统管理学报, 2018, 27(5): 854-862.Shi Baofeng, Wang Jing. A credit rating model of microfinance for farmers based on ELECTRE III［J］. Journal of Systems & Management, 2018, 27(5): 854-862.
［27］林宇, 黄迅, 淳伟德, 等. 基于ODR -ADASYN-SVM的极端金融风险预警研究［J］. 管理科学学报, 2016, 19(5): 87-101.Lin Yu, Huang Xun, Chun Weide, et al. Early warning for extremely financial risks based on ODR-ADASYN-SVM［J］ Journal of Management Sciences in China, 2016, 19(5): 87-101.
［28］ Chawla N V, Bowyer K W, Hall L O, et al. SMOTE: Synthetic minority over-sampling technique［J］. Journal of Artificial Intelligence Research, 2002, 16(1): 321-357.
［29］杨莲, 石宝峰, 迟国泰, 等. 非均衡数据下基于BPNN-LDAMCE的信用评级模型设计及应用［J］. 数量经济技术经济研究, 2022, 39(3):152-169.Yang Lian, Shi Baofeng, Chi Guotai, et al. Design and application of a credit rating model based on BPNN-LDAMCE with imbalanced data［J］. The Journal of Quantitative & Technical Economics, 2022, 39(3): 152-169.
［30］ Barry M J A, Linoff G S. Data mining techniques: For marketing, sales, and customer support［M］. New York: John Wiley & Sons, 1997.
［31］ Glorot X, Bordes A, Bengio Y. Deep sparse rectifier neural networks［J］. Journal of Machine Learning Research, 2011, 15: 315-323.
［32］杨莲, 石宝峰, 董轶哲. 基于Class Balanced Loss修正交叉熵的非均衡样本信用风险评价模型［J］.系统管理学报,2022,31(2):255-269,289.Yang Lian, Shi Baofeng, Dong Yizhe. A credit risk evaluation model for imbalanced data classification based on class balanced loss modified cross entropy function［J］ Journal of Systems & Management, 2022, 31(2):255-269,289.
［33］ Chi Guotai, Abedin M Z, Moula F. Modeling credit approval data with neural networks: An experimental investigation and optimization［J］. Journal of Business Economics and Management, 2017, 18(2): 224-240.
［34］ He Hongliang, Zhang Wenyu, Zhang Shuai. A novel ensemble method for credit scoring: Adaption of different imbalance ratios［J］. Expert Systems with Application, 2018, 98: 105-117.
［35］ Wang Di, Zhang Zuoquan, Bai Rongquan, et al. A hybrid system with filter approach and multiple population genetic algorithm for feature selection in credit scoring［J］. Journal of Computational and Applied Mathematics, 2017, 329: 307-321.

基于Focal Loss修正交叉熵损失函数的信用风险评价模型及实证

Credit Risk Evaluation Model and Empirical Research Based on Focal Loss Modified Cross-Entropy Loss Function

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 1

Metrics

本文评价

推荐阅读 0