融合网络结构特征的贷款违约预测研究  

Research on Loan Default Prediction Based on Social Network and Feature Mining Model

作　　者：孙玮 刘东琪 靳晓曼 SUN Wei;LIU Dongqi;JIN Xiaoman(School of Finance,Hebei University of Economics and Business,Shijiazhuang,China,050000)

机构地区：[1]河北经贸大学金融学院,石家庄050000

出　　处：《福建电脑》2024年第10期18-22,共5页Journal of Fujian Computer

基　　金：河北经贸大学项目“基于第三方量化平台的策略开发实验项目”(No.2021JYQ02);“生成式人工智能模型纳入股票市场信息环境的潜在风险与应用对策研究”(No.2024ZD01);河北省金融学会项目“数字人民币生态系统研究:要素构成、圈层交互与协同演化”(No.xh2024036);河北省人力资源社会保障科研合作项目“生成式人工智能在劳动力市场中的影响研究——基于河北省的扎根理论分析”(No.JRSHZ-2024-02016)资助。

摘　　要：信贷违约风险是金融风险的重要组成部分,应用机器学习进行违约预测已成为研究重点。为提升机器学习模型在贷款违约中的预测能力,本文构建了贷款用户的社会网络。通过选取体现关联关系的网络结构特征,并应用DeepWalk算法和Stacking模型,将用户的社会网络拓扑结构信息作为特征加入机器学习模型进行训练。实验结果表明,该模型能够提高预测准确度,较基准模型AUC分别提高了1.38%、1.74%,对金融机构在贷前风险识别及制定信贷决策具有借鉴意义。Credit default risk is an important component of financial risk,and the application of machine learning for default prediction has become a research focus.To enhance the predictive ability of machine learning models in loan defaults,this paper constructs a social network of loan users.By selecting network structure features that reflect correlation relationships and applying DeepWalk algorithm and Stacking model,the user's social network topology information is added as a feature to the machine learning model for training.The experimental results show that the model can improve prediction accuracy,with AUC increased by 1.38%and 1.74%respectively compared to the benchmark model.This method has reference significance for financial institutions in identifying pre loan risks and making credit decisions.

关 键 词：社会网络 DeepWalk算法 Stacking模型 贷款违约 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]