Research On Loan Default Risk Identification Based On Combination Forecast Model

In recent years,with the rapid development of China’s economy,a large number of online loan platforms have emerged.At the same time,the credit risk has become increasingly serious,and many defaults and frauds of the borrower have resulted in a serious influence on the whole financial field.Therefore,how to prevent defaults and identify potential default borrowers plays a vital role in the survival and development of such platforms.Basing on the loan transaction information data for two quarters provided by Lending Club,this paper selects the information data of the platform’s borrowers for research.With the help of Python and R,the data is processed and used to build a portfolio prediction model for loan risk identification.The specific research contents are as follows:(1)Firstly,the data are filtered,the missing values are processed and the feature codes are processed,and the factors that affect the default of the loan are analyzed.That is,the profile analysis of online loan default users.The research shows that the risk of default is higher for the borrower with higher loan amount,so the platform should evaluate the risk of default more carefully.Also,The working year of borrowers is an important criterion for assessing the risk of default,the longer the working years of the borrower,the lower the risk of default.borrowers of leased house face greater economic pressure,which will lead to a higher risk of default.(2)The number of defaulters and non-defaulters is quite different in the data set in this paper.For the unbalanced data set,the method combined oversampling with undersampling will be used,and the majority of samples will be under-sampled by random sampling,the few samples will be over-sampled by Borderline-SMOTE.Then filter and wrapper methods are used to filter out the features and to construct multiple individual learner models and stacking fusion model.(3)The probability of first type error is taken as a penalty parameter of individual learners’ s weight,and the Lagrange multiplier is used to obtain the optimal solution of weight,so as to construct the combined forecasting model.The combined model based on the method of linear weighting to combine multiple individual learners,which can utilize the advantages of each individual learner and improve the classification prediction effect of the model.It also can reduce the probability of first type error to great extent.This model can effectively help the online loan platform to identify the default risk and reduce the loss,which is of great significance to improve the healthy development of online loan platforms.(4)According to the results,this paper provides some suggestions for the online loan platform in establishing a reasonable and effective risk prevention and control system,completing the information sharing of various industries and optimizing the personal credit system.