Credit Risk Forecast Based On IF And LSTM

With the development of Internet finance,consumers seem increasingly open to overconsuming.In recent years,consumption credit and loans is growing by leaps and bounds in China.Only customers’ personal information is involved in traditional credit risk prediction models,which ignore the impact of changes in the macro environment on the possibility of customers to repay on time.This is confirmed by the extremely irregular overdue rate during the COVID-19 pandemic and the low prediction accuracy of traditional models at that time.Therefore,it is very necessary to build an effective credit risk prediction model due to the early stage of consumption credit and loans and the existing relatively primitive risk control models.This thesis is an exploratory study on the impact of the macro environment on the overdue rate.The author proposes a new credit risk prediction model based on Logistic regression,Iterative Filtering and LSTM network.The specific steps of this model are as follows.Firstly,the preliminary predicted value of the overdue rate is obtained according to the constructed Logistic regression model.Secondly,Iterative Filtering and LSTM network are used to measure the impact of the macro environment.Finally,coming out with the final result after the preliminary predicted value of the overdue rate is corrected.And in the process of preprocessing and feature extraction of the macro-environmental factors,a new convergent Iterative Filtering method—Exponential Iterative Filtering method is proposed,which decomposes the non-stationary sequence into sub-sequences,that contain the trend and characteristics of the original signal.In this thesis a variety of other prediction models are established: Logistic regression model,LG-LSTM model and Random Forest model.Through comparative experiments on various datasets,the proposed model achieves the highest prediction accuracy compared with other models.In addition,after testing with the dataset during the COVID-19 pandemic,it is found that the proposed model can measure the impact of emergencies on the overdue rate,and has high prediction accuracy compared with the poor performance of traditional models during this period.The credit risk prediction model based on LG,IF and LSTM quantifies the impact of the macro environment on credit risk,and provides a brand new idea for credit risk forecast research.