Application Research Of Internet Financial Credit Risk Control Model Based On Machine Learning

With the rise of the concept of "Internet plus",the traditional financial industry has transformed to Internet finance.As the core of Internet finance,personal credit business has rapidly increased its market share by virtue of its low threshold,convenient operation and no intermediary.However,there are problems in the field of credit finance,such as huge data,business complexity,and difficulty in balancing income and risk.Therefore,establishing a credit risk control model with accurate credit evaluation ability,strong business interpretation and stable decision-making output is the core competitiveness of credit financial institutions.Aiming at the credit evaluation of the financial business related to credit financial institutions,based on the real personal credit loan data set of a credit financial institution,Python is used as a tool for data analysis and model construction.Starting with data analysis,feature screening,feature engineering,machine learning model construction,and model effect evaluation,combining the professional knowledge of credit financial risk control,the personal credit score card model is finally established.This model provide methods and suggestions with reference significance for financial institutions.The main research contents of this paper are as follows:(1)Data preprocessing and feature engineering.Use real personal credit loan data set for research and analysis.First,Smote oversampling algorithm is used to balance the proportion of positive and negative samples for the sub-balanced data set.Then the features are screened based on the IV value,Pearson correlation coefficient and VIF variance inflation factor.Carry out feature engineering such as chi-square sub-box processing and WOE coding for the screened features.Finally,PSI is used to test the stability of the feature.(2)Build a credit risk control model.By introducing machine learning algorithms such as logical regression,random forest and XGBoost,and using grid search method to adjust parameters,a credit risk control model is constructed.Select the Accuracy,f1-score,KS value,PSI and AUC value to evaluate the effect of the model,and compare and analyze the advantages and disadvantages of different models based on the actual business of credit risk control.The experiment found that the model based on the split WOE has a good performance in the field of credit finance,among which the logical regression has a stronger ability to classify and identify users.(3)The credit score card is constructed based on the logical regression model.The empirical analysis shows that the credit score card has a good ranking for the user’s credit prediction,and can accurately determine the customer group according to the credit score,and the logical regression credit score card can accurately output the credit score of each feature.The risk control decision system integrated with logic regression card and XGBoost can effectively evaluate the credit risk status of users,enrich the landing scenarios of credit finance business,and meet the needs of credit finance related institutions to balance income and risk.This paper adopts data analysis algorithms and machine learning models that are suitable for the actual business scenarios of credit finance,which can better predict the default risk of users’ credit finance business,and establish a user credit evaluation system that can meet the business needs,providing a reliable basis for the credit business decisions of relevant institutions.