The Study Of Credit Risk Prediction Based On Ensemble Learning

Consumer credit is a kind of credit activity that financial institutions provide loan services to meet consumer demand for goods and services and require users to repay on time.With the rapid development of Internet finance,credit risk is also a major problem faced by many financial institutions.Credit risk that the user does not fulfill the debt after the bank provides financial services to the user.Credit risk has always been the main research area in bank loan decision-making.This paper mainly studies the application of ensemble learning algorithms in credit risk prediction.The article introduces simple machine learning algorithms:KNN,Naive Bayes,Logistic Regression,Decision Tree.Ensemble Learning Algorithm-s:GBDT,XGboost,Random Forest are three kinds of algorithms with decision tree as the base learner and GBM with random forest as the base learner,as well as some indicators to evaluate the performance of the model.In empirical analysis,the feature is dumbed at first.Then the data is fitted by the simple machine learning algorithm.The optimal model—Decision Tree is selected using the model evaluation indicators such as precision,recall and AUC.Finally,the decision tree is used as a base learner for further ensemble learning.We found that the fitting effect of GBM is the best and the effect of random forest is second,which provides a reference for the credit approval of financial institutions' loan business.At present,the application of GBM in China mainly focuses on traffic and pedestrian detection,e-commerce,anomaly detection,etc..There are even fewer applications in other fields,especially GBM with random forest as the base learner.The innovation of this paper is to introduce GBM into the prediction of credit riskj,using GBM with random forest as the base learner and using precision and recall as indicators for model evaluation.