Research On Credit Loan Default Prediction Based On Migration Learning

In recent years,China’s rapid development of the digital economy,so that the commercialization of financial scenarios are increasingly rich,the original only traditional banks to provide credit loans and other services become more inclusive,more and more individuals and small and micro enterprises can easily obtain credit loans and other financial services in the Internet financial platform,and with the exponential growth of the credit loan business scale,banks and financial platforms face credit risk is also The credit risk faced by banks and financial platforms is increasing day by day.Traditional credit risk control models rely on manual evaluation and empirical judgment,which are inefficient and imprecise in risk control,while the accuracy of credit default prediction has been greatly improved with the application of machine learning models in the credit risk control field in recent years.In order to solve this problem,this paper proposes a migration learning-based credit default prediction scheme to improve the prediction accuracy of the model and help credit institutions strengthen risk control by migrating external data,features,and models,with the following main research work:(1)The personal credit data of ZY Bank is selected as the initial data,exploratory analysis of credit user information is performed,and data preprocessing,including missing value processing outlier processing and variable coding,and five default prediction classifiers based on machine learning methods such as logistic regression,random forest,and integrated learning are established,and the experimental results show that the basic model in default prediction for a single data set The experimental results show that the base models can achieve better results in default prediction for a single dataset,and these models are used as comparison models for migration learning schemes.(2)The online credit data of ZY bank is selected as external data to build a migration learning based credit default prediction,an innovative external data filter based on Catboost framework is constructed,and derived features are constructed based on credit business knowledge and machine learning method of K-means clustering,and finally the external data and features are merged with the initial data selection,and finally the Lightgbm model as a classifier to realize the migration process of data,features,and models,and the AUC value of the migration learning scheme reaches 0.883,which is significantly higher than other models.