Research On Personal Credit Loan Risk Control

In recent years,the trend of commercial banking business towards retail de-velopment has become increasingly apparent,which makes the credit operations become an important components of the development of banking.Personal credit loans have become an important fulcrum of banking business and the demand is constantly growing.In the "Internet+" environment,personal credit business has been everywhere in daily life,including the small daily necessities,cars,hous-es and so on.The vigorous development of personal credit business has become one of the main measures to remove capital constraints and improve profitability.However,the rapid development of the credit economy has also brought cred-it loan risks.Therefore,it is extremely important in real life to study the risk of users'loan default,assess the risk level,and do a good job of risk control.From the perspective of machine learning and integrated learning,this article uses empirical research on the real data of personal credit loans published abroad.In terms of data processing,this paper first uses the filtration method to select the influencing features,and then uses the SMOTE algorithm to deal with the sample imbalance between the default sample and the fufillment sample.In terms of empirical analysis,this paper uses the three models of Logistic Regression,Naive Bayes,and Random Forest to predict the personal credit data.In evaluating the training effect of the model,the six indexes of accuracy rate,accuracy rate,recall rate,FI value,AUC value and ROC curve are synthetically used.The training effects of the three models are analyzed and compared,and it is found that Random Forest is more advantageous than the other two models in personal credit risk prediction.When establishing the risk control model of personal credit,this paper adopts the Boruta algorithm in feature selection,finally,15 features are retained,in-cluding the nature of work,percentage of installment payment,living time in the current residence,personal assets,age,account balance,other installment payments plans,the history performance of credit,housing status,whether there is a guarantor,duration of account balance status,credit purpose,loan amount,personal creditworthiness and working time.Then we predict the sample data set after it is selected again according to the 15 features,and find that the Boruta-Random Forest is better than the Random Forest.Finally,we transform the predicted default probability into personal credit scores,and establish a model of personal credit risk control.This paper uses multiple machine learning models to evaluate personal credit,and the innovations are as follows:1,In the application of Random Forest model to personal credit evaluation,Boruta algorithm is combined to select effective features and optimize the method of feature selection.After selecting the effective features,this paper uses the Boruta-random forest model to predict the personal credit data set,and the prediction results are more accurate than before.2.When constructing the model of personal credit risk control,this paper adopts a method of transforming the default probability into a credit score.Whether the borrower's loan application can be passed or not is determined according to the grade of the scores,which ensures the feasibility of the model in practical applications.