非均衡多分类绿色信贷信用风险预测

打开文本图片集

中图分类号：F832.4 文献标识码：A 文章编号:1674-0033(2025)06-0064-11

¹，1²，1

(1. ，;2.宁夏科学计算与智能信息处理协同创新中心，宁夏银川 750021)

Imbalanced Multi-class Classification for Green Credit Risk Prediction

FU Hong-xin1,GAO Yue-lin¹²,HAN Zhi-xin1 (1.Mathematicsand the Information Science College,North Minzu University,Yinchuan75Oo21,Ningxia; 2.Ningxia Collaborative Innovation Center of Scientific Computing and Intelligent Information Procesing, North Minzu University, Yinchuan 750021,Ningxia)

Abstract: To addressthe imbalance in green credit data and the critical role of multi-level credit prediction in pre-loan risk control,an OVO-ISMOTENN-RF hybrid model is proposed for risk assessment. This model breaks down the multi-class problem using the One vs One strategy,and it employs the ISMOTENN algorithm to optimize the minority class samples,helping to alleviate data imbalance.Additionally，it combines a random forest classifier optimized by particle swarm optimization,with final resultsgenerated through majority voting.The empirical results show strong performance: the accuracy rates for the low-risk and highrisk categories are O.79 and O.78,respectively,with an overallaccuracy of O.66 and an M-R score of 0.63. SHAP analysis highlights the importance of variables such as R&D investment and working capital turnover rate,providing valuable insights for the bank's risk management strategies.

KeyWords: imbalanced multi-classclassification; green credit; ISMOENN algorithm; SHAP algorithm

机器学习万法因具不依赖数据分布假设，能高效处理海量数据，成为信贷风险预测的重要手段。（剩余16052字）