Research And Application Of Loan Default Prediction Model Based On XGBoost-Stacking Ensemble Learning

Accurately identifying potential default loans,preventing loan issuance in advance,or tracking the loans that have been issued to ensure loan returns are the core issues in the field of financial risk control.Aiming at the problems of low classification performance,lack of stability and insufficient generalization ability of the current loan default prediction model,this paper uses the financial loan data set to construct a loan default prediction model based on GSCV-XGBoost,and proposes a loan default based on XGBoost-Stacking integrated learning Prediction model,develops a loan default prediction system based on the Django framework.The main research work and achievements include the following:1.A loan default prediction model based on GSCV-XGBoost is constructed.First,the financial loan data set is preprocessed,combined with descriptive statistical analysis to measure the correlation between feature variables and target variables,and feature subsets are selected to construct a loan default prediction model based on XGBoost.Secondly,we use grid search and five-fold cross-validation methods to optimize the model parameters and train the model to form a loan default prediction model based on GSCV-XGBoost.The experimental results show that the GSCVXGBoost model performs better in classification effect and performance than the four models of Support Vector Machine,K-Nearest Neighbor,Logistic Regression,and Random Forest after parameter optimization.2.We propose a loan default prediction model based on XGBoost-Stacking ensemble learning.We adopt the idea of multi-model fusion of Stacking integrated learning algorithm,and use logistic regression,random forest,and K-nearest neighbor models as the first layer of base learners of Stacking algorithm,and use XGBoost-based loan default prediction model as the second layer of base learners,so as to overcome the defects of single model through model layered fusion and improve the classification performance,stability and generalization ability of loan default prediction.The model is compared with four single models.Experiments show that compared with the single model,this model has better classification performance,stability and generalization ability.3.Based on the loan default prediction model constructed in 1.2,a loan default prediction system is developed using the Django framework to achieve effective prediction and analysis of loan defaults.Main contributions: Based on the financial loan dataset,a loan default prediction model based on GSCV-XGBoost is constructed by combining grid search and five-fold cross-validation methods;a multi-model hierarchical fusion is performed using Stacking algorithm,and a loan default prediction model based on XGBoost-Stacking integrated learning with better performance is proposed;A loan default prediction system is developed to achieve effective prediction and analysis of loan defaults.