Research On The Prediction And Application Of The Overdue Risk Of Mutual Money Loan Based On Penalty Regression

With the rapid development of the emerging internet finance industry and the implementation of regulatory policies,the credit information system has been continuously improved and the online lending platform has faced credit risk issues in the HP financial process.This article uses the variable selection method of penalized regression to apply to Internet finance overdue loan risk prediction.The first chapter introduces the basics of internet finance credit information,loans,financial technology companies,laws and regulations,and network information security.It also introduces the application of R language in the Linux environment,highlights the research progress of the two classification problems,and points out that supervised learning approaches face generalization,overfitting,and the curse of dimensionality.The second chapter introduces linear model theory,generalized linear model theory,regression model with penalty terms,such as lasso penalty of L1 norm,elastic-net penalty of L1 and L2 norm,ridge penalty of L2 norm,SCAD penalty,etc.,Combining multivariate normal distribution random number simulation results to compare coefficient estimates for various methods.The third chapter introduces the performance evaluation method of the classifier.The R language is used to implement the calculation of each index of the confusion matrix,non-parametric McNemar test and Kolmogorov–Smirnov test.The numerical simulation content is combined with the logistic regression prediction values in Chapter 2 to calculate the optimal accuracy and the optimal threshold.The fourth chapter is an example of data analysis.Using R language to process overdue data,combined with kernel density estimation and other methods to conduct exploratory data analysis on loan data in the Internet finance field,including logistic regression,lasso penalized regression,elastic-net(?=0.5)penalized regression,ridge regression,etc.To construct overdue risk prediction models to achieve classification.Forecast and calculate the probability of overdue.Comparing the accuracy of varioustypes of models,recall rates,AUC indicators,chi-square statistics,D values of Kolmogorov–Smirnov test,etc.,statistical tests show that the penalized regression can distinguish between overdue and not overdue samples.In addition,different McNemar tests and comparisons of accuracy were performed for different models at different thresholds to compare model differences.The application results show that lasso penalized regression for overdue risk warnings and has application value in the field of Internet finance risk control.