主管:中国科学院
主办:中国优选法统筹法与经济数学研究会
   中国科学院科技战略咨询研究院

中国管理科学 ›› 2026, Vol. 34 ›› Issue (11): 120-129.doi: 10.16381/j.cnki.issn1003-207x.2023.2229cstr: 32146.14.j.cnki.issn1003-207x.2023.2229

• • 上一篇    下一篇

基于自适应BPNN-ACPL非均衡信用评价模型及实证

杨莲1, 刘文秀2, 柴娜娜3,4, 石宝峰3,4()   

  1. 1.山东农业大学经济管理学院,山东 泰安 271018
    2.南京师范大学商学院,江苏 南京 210023
    3.西北农林科技大学经济管理学院,陕西 杨凌 712100
    4.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:2023-12-30 修回日期:2025-08-22 出版日期:2026-11-25 发布日期:2026-11-10
  • 通讯作者: 石宝峰 E-mail:shibaofeng@nwsuaf.edu.cn
  • 基金资助:
    国家社会科学基金重大项目(23&ZD175)

Credit Evaluation Model and Empirical for Imbalanced Data Based on Adaptive BPNN-ACPL

Lian Yang1, Wenxiu Liu2, Nana Chai3,4, Baofeng Shi3,4()   

  1. 1.College of Economics and Management,Shandong Agricultural University,Taian 271018,China
    2.Business School,Nanjing Normal University,Nanjing 210023,China
    3.College of Economics and Management,Northwest A & F University,Yangling 712100,China
    4.Research Center on Credit and Big Data Analytics,Northwest A & F University,Yangling 712100,China
  • Received:2023-12-30 Revised:2025-08-22 Online:2026-11-25 Published:2026-11-10
  • Contact: Baofeng Shi E-mail:shibaofeng@nwsuaf.edu.cn

摘要:

针对以交叉熵为目标函数的反向传播神经网络模型存在对非均衡样本适用性不强、尤其对金融机构重点关注的违约样本识别精度不高的现象,将图像识别领域ACPL(adaptive class perception loss)函数引入信用风险评价,构建对样本权重具有自适应调节功能的BPNN-ACPL (back propagation neural network-adaptive class perception loss)非均衡信用评价模型。采用3个真实信贷数据、7种对比模型、5种模型评价准则、2种不同训练集和测试集划分比例,对所建模型信用风险预测性能和稳健性进行评估。实证结果表明:相比较基于样本损失重加权思路构建的BPNN-CBCE (back propagation neural network-class balanced cross entropy)非均衡信用评价模型,本文所提模型在提升违约样本识别力、避免对非违约样本误判方面具有优势;相比较传统反向传播神经网络、多种浅层机器学习模型,本文所提模型在识别违约客户、有效区分两类客户方面效果明显;本文所提模型在多个信贷数据集中表现出稳健的信用风险预测性能。创新与特色:在交叉熵函数中引入可以自适应调节两类样本损失的调节项,可以根据样本违约预测值判断样本是否被误判,进而自适应计算适合样本被准确识别的梯度,当样本被误判时,同时保留样本为自身累积的正梯度以及施加给对立类样本的负梯度;当样本被准确判别时,仅保留样本为自身累积正梯度。引入调节项后的ACPL函数既可提升模型对违约样本的学习力度和预测精度,还能避免对非违约样本的误判,为非均衡样本信用风险评价提供新的参考。

关键词: 信用评价, 非均衡, ACPL损失, 交叉熵

Abstract:

In response to the weak applicability of backpropagation neural network models with Cross Entropy as the objective function to imbalanced samples, especially the low recognition accuracy of default samples that financial institutions focus on, the ACPL (Adaptive Class Perception Loss) function in the field of image recognition is introduced into credit risk assessment to construct a BPNN-ACPL (Back Propagation Neural Network-Adaptive Class Perception Loss) imbalanced credit evaluation model with adaptive adjustment function for sample loss weights.In order to evaluate the performance and robustness of the credit risk prediction model, three real credit data, seven comparative models, five model evaluation criteria, and two different training and testing set partitioning ratios are used. The three empirical data are 1298 farmer data from a small credit institution in China, 2044 farmer data from a large commercial bank in China, and 1000 German credit data publicly released by UCI. The seven comparative models are BPNN-CBCE (Back Propagation Neural Network-Class Balanced Cross Entropy), BPNN-CE (Back Propagation Neural Network-Cross Entropy), XGBoost (eXtreme Gradient Boosting), SVM (Support Vector Machine), DT (Decision Tree), RF (Random Forest), and KNN (K-Nearest Neighbors). The five evaluation criteria are ACC (Accuracy), AUC (Area under the ROC curve), NDR (Non-default recall), DR (Default recall), and GM (Geometric mean). The ratio of dividing the two different training and testing sets is 9∶1 and 7∶3, respectively. Empirical findings show that compared to the BPNN-CBCE model constructed based on sample loss reweighting, the proposed model has advantages in improving the recognition of default samples and avoiding misjudgment of non-default samples; Compared with traditional backpropagation neural network models and shallow machine learning models, the proposed model has significant effects in identifying defaulting customers and effectively distinguishing between the two types of customers; The proposed model demonstrates robust credit risk prediction performance on multiple credit datasets. Introducing an adjustment term in the Cross Entropy function that can adaptively adjust the loss of two types of samples. Its essence is to determine whether the sample has been misjudged based on the predicted value of sample default, and then adaptively calculate the gradient suitable for accurate identification of the sample. When the sample is misjudged, it retains both the positive gradient accumulated by itself and the negative gradient applied to the opposite class sample. When the sample is accurately identified, only the sample is retained as its own accumulated positive gradient. The introduction of the adjustment term in the ACPL function can not only improve the learning strength and prediction accuracy of the model for default samples, but also avoid misjudgment of non-default samples, providing a new reference for credit risk evaluation of non-equilibrium samples.

Key words: credit evaluation, imbalance, ACPL loss, cross entropy

中图分类号: