农业经济问题Issue(11)：26-45,20.
基于样本优化与可解释机器学习的农户信用风险智能评估研究
Research on Intelligent Assessment of Farmer Credit Risk Based on Sample Optimization and Interpretable Machine Learning
摘要
Abstract
Against the backdrop of the accelerated digital transformation of rural finance,accurately assessing the credit risk of farmers has become a key to promoting the healthy development of rural fi-nancial market and advancing inclusive finance.However,existing research still faces many challenges in dealing with the sample imbalance of rural credit risk,improving classification accuracy,and enhan-cing model interpretability.To address these issues,this paper integrates MHLS with the ability to opti-mize sample balance,ALR-LightGBM with an adaptive learning rate adjustment mechanism,and SHAP for interpretability,to construct a new MHLS-ALR-LightGBM-SHAP credit risk assessment mod-el.Based on the real farmer credit data from the Rural Credit Union of Fujian Province,this study clas-sifies farmers into three categories:"normal","concern",and"default",and comprehensively uses over 20 key farmer characteristic indicators.By comparing with seven benchmark models including RF,DT,KNN,GNB,LR,SVM,and BPNN,the effectiveness of the MHLS-ALR-LightGBM-SHAP model is verified,and the global and local explanations of the influence of important features are achieved through the SHAP method.The research results show that:(1)The proposed model in this study has higher classification accuracy and stronger minority class recognition ability compared with the bench-mark models.(2)Evaluation indicators such as no principal repayment and loan renewal,credit rating,and recent overdue status are key features for default discrimination.(3)There are significant heteroge-neous effects of different regions on farmer credit risk,with farmers in coastal areas of Fujian(such as Putian and Quanzhou)having relatively good credit conditions,demonstrating a strong economic founda-tion and a low default tendency,while farmers in mountainous areas of Fujian(such as Sanming and Nanping)show a higher default tendency.The research results provide a risk identification tool that is both accurate and interpretable for rural financial credit risk research,which is of great significance for improving the efficiency of credit risk management and promoting the high-quality development of rural finance.关键词
农户信用风险/风险识别/ALR-LightGBM/可解释机器学习Key words
Farmer credit risk/Risk identification/ALR-LightGBM/Explainable machine learning引用本文复制引用
汪寿阳,张熠,张婷婷,郑海荣..基于样本优化与可解释机器学习的农户信用风险智能评估研究[J].农业经济问题,2025,(11):26-45,20.基金项目
国家社会科学基金一般项目(编号:20BJY153),福建省社会科学研究基地重大项目(编号:FJ2023JDZ030,FJ2022JDZ020),福建省财政补助资金项目(编号:KLY24107XA,KSC22R01A). （编号:20BJY153）
