基于模型融合的贷款违约预测模型研究

Research on Loan Default Prediction Model Based on Model Fusion

在银行贷款服务中,用户的违约行为对于银行来说存在着信用风险损失。用户的相关指标与是否会发生违约行为有着一定关系。针对贷款违约问题,文章提出基于模型融合的贷款违约预测模型,通过预测模型预测用户违约情况,从而降低损失风险。文章基于天池贷款违约预测大赛的数据集,对其数据进行数据分析、处理及数据类型转换等工作,确定违约相关的主要特征,包括贷款金额、贷款期限、贷款利率、分期付款金额、贷款等级、就业年限、年收入、债务收入比等。基于对各类模型的比较,文章选取XGBoost和LightGBM算法,引入Focal Loss损失函数,通过Stacking方法进行融合,搭建出FL-XGBoost-LightGBM违约预测模型。实验表明,引入Focal Loss损失函数的模型精度优于未引入损失函数模型精度;引入Focal Loss损失函数的融合后的模型精度比单一模型的精度高,取得最好的预测结果。

In bank loan services,there is a credit risk loss for the bank for the user*s default.The user's metrics have a relationship with whether a default will occur.Aiming at the problem of loan default,this paper proposes a loan default prediction model based on model fusion,which predicts user default through the prediction model,thereby reducing the risk of loss.Based on the dataset of the Tianchi Loan Default Prediction Competition,this paper analyzes,processes and converts the data to determine the main characteristics related to default,including loan amount,loan term,loan interest rate,installment amount,loan grade,employment period,annual income,debt-to-income ratio,etc.Based on the comparison of various models,this paper selects XGBoost and LightGBM algorithms,introduces the Focal Loss loss function,and integrates them through the stacking method to construct the FL-XGB oost-LightGBM default prediction model.Experiments show that the accuracy of the model with Focal loss loss function is better than that of the model without loss function.The accuracy of the fused model with Focal Loss loss function is higher than that of a single model,and the best prediction results are obtained.