【摘要】 首先分析了互联网消费金融违约风险的特征，运用信息经济学研究互联网消费金融借款人违约风险的形成机制，构建了借款人信用风险评价指标体系。然后构建集XGBoost、LightGBM和CatBoost的二分类加权硬投票融合模型，用于预测借款人违约风险。通过实证检验比较不同分类器训练下的预测结果发现：融合模型的预测精准度优于单一模型XGBoost、LightGBM和CatBoost;基于特征重要性排序结果，发现影响贷款者违约的关键因素主要包括贷款利率、年收入、公共事业差评数记录、循环信贷余额等。最后提出了加强借款人违约风险管理的建议。更多还原

【Abstract】 This paper first analyzes the characteristics ofInternet consumer finance default risk, studies the formation mechanism of Internet consumer finance borrower default risk by using information economics, and constructs the Borrower Credit Risk Evaluation Index System and two-class weighted hard voting fusion model.This model is composed of Xgboost, LightGBM and CatBoost to predict the default risk of borrowers.It is found that the fusion model is superior to the single model XGBoost, LightGBM and CatBoost in prediction accuracy, and that the key factors affecting the borrower’s default mainly include loan interest rate, annual income, bad rating record of public utilities, revolving credit balance and so on.Finally, the paper puts forward some suggestions on how to strengthen default risk management of internet consumer finance platform.更多还原