Construction And Application Of Internet Financial Loan Default Prediction Model Based On Stacking Fusion Algorithm

With the continuous development of Internet technology,a new online loan business relying on Internet technology has emerged.Compared with the loan business of traditional banks,the online lending business realized through the Internet has the characteristics of lower threshold and more convenient business realization.However,at the same time,due to the low threshold of Internet loans,it is inevitably accompanied by the high risk of investment income,and when the relevant national regulatory policies were not yet perfected in the early stage,many Internets financial untrustworthy incidents occurred.Therefore,to predict whether the borrower will default in the future,the machine learning algorithm can be used to deal with this classic binary classification problem.By training a large number of loan defaulting users’ data to obtain an algorithm model,it can more accurately and efficiently identify users who may default on loans,thereby reducing the operational risks of Internet financial platforms and helping the Internet finance industry develop healthily and in the long run.In this thesis,the Stacking fusion algorithm is used to construct the Internet financial loan model,the Alibaba Cloud Tian chi dataset is used as the training and test set of this thesis,the KNN algorithm,Logistic regression,random forest,XGBoost,Ada Boost and Light GBM algorithms are used as the base models to build the Stacking fusion model,and the five-fold cross-validation method and Bayesian parameter optimization method are used to determine the optimal parameter combination.The four model evaluation indicators of F1 value and AUC value were evaluated for model classification effect,and finally logistic regression was deleted,and the random forest,XGBoost,Ada Boost and Light GBM algorithms were used as the first layer model,and the KNN algorithm was used as the meta-model to build the Stacking fusion model,and the Stacking fusion model was better than the single model in the four model evaluation indicators.At present,China’s Internet financial market has entered a stage of stable development,but enterprises with Internet financial business still need to improve the ability to review borrowers,the model built in this thesis can be used in the actual borrower audit system,while adding more borrower behavior information to improve the accuracy of model classification,and this model can also be applied to the personal credit system,which is expected to provide theoretical guidance and help for China’s credit investigation department for the supervision of the Internet finance industry.