基于FPFF-Blending模型融合的个体工商户信用评价研究  

Research on Credit Evaluation of Individual Businesses Based on FPFF-Blending Model Fusion

作　　者：任军霞 陈瑞勇 叶宇轩 孙秀文 唐嘉成 李响 Ren Junxia;Chen Ruiyong;Ye Yuxuan;Sun Xiuwen;Tang Jiacheng;Li Xiang(Zhejiang Zhelixin Credit Information Co.,Ltd,Hangzhou 310000,Zhejiang,China)

机构地区：[1]浙江浙里信征信有限公司,浙江杭州310000

出　　处：《征信》2023年第4期64-71,共8页Credit Reference

摘　　要：个体工商户信用评价研究往往通过单一机器学习模型建立,其预测精确率较低,抗干扰能力较弱。基于特征金字塔的FPFF特征融合算法,应用于Blending模型融合框架,建立个体工商户信用评价异质融合模型,并赋予模型可解释性,综合解决单一模型稳定性较差、原有Blending框架融合模型过拟合、融合模型缺乏可解释性的问题。通过对个体工商户数据集进行实证实验,结果表明:融合模型较单一机器学习模型在个体工商户信用评价场景下具有更优的预测性能和泛化能力。The research on credit evaluation of individual businesses is often established by a single machine learning model,which has low prediction accuracy and weak anti-interference ability.The FPFF feature fusion algorithm based on feature pyramid is applied to the Blending model fusion framework to establish a heterogeneous fusion model for credit evaluation of individual businesses,and endow the model with interpretability,comprehensively solving the problems of poor stability of a single model,overfitting of the original Blending framework fusion model,and lack of interpretability of the fusion model.Through empirical experiments on individual business data sets,the results show that the fusion model has better prediction performance and generalization ability than the single machine learning model in the individual business credit evaluation scenario.

关 键 词：个体工商户 信用评价 特征金字塔 FPFF特征融合算法 Blending融合框架 SHAP可解释性 

分 类 号：F830[经济管理—金融学] F276