Cash Flow Forecast Of Online Loan Platform Based On LSTM Structure

With the development of economy and Internet,network lending emerges as the times require and gradually takes shape.It has the characteristics of convenient financing and wide range of objects.It can provide convenient financing services for small and medium-sized enterprises and ordinary people,and it is an indispensable part of the national Inclusive Finance.Because of this,it is necessary for us to study the online lending platform to help it develop in a more healthy direction.In the related research of online loan,in addition to the regulatory issues controlled by the state,the most important one is the risk management of online loan platform itself,in which it is particularly important to timely grasp the repayment situation of borrowers.In the past,the main focus of research is to predict the credit risk of the repayment,that is,to predict whether the repayment will default when due.Once default,online loan platform is more violent collection,which is not conducive to the sound development of the platform,and the prediction of users' credit risk is not conducive to the better development of the platform.Considering that the users of online loan platform are mostly small loan users,when the number of users is large and the amount of loan is different,the management of cash flow is easy to encounter difficulties.Only by solving the problem of cash flow management of online loan platform,can we realize the virtuous cycle of loan and loan,so as to truly help the healthy development of online loan platform.On the basis of previous research on credit risk,this paper further studies the cash flow of the online loan platform,including the repayment date and amount of users,which can help the platform to better manage its funds.First of all,this paper determines that the research content is to predict the user's repayment in the last period.Since the independent variable includes not only the variable that does not change with time,but also the variable that changes with time,the model based on time series data and the model based on section data are established respectively to predict the repayment date and the repayment amount of the user in the last period.33 cases(0-31 days in advance and non repayment)represent the predicted repayment date,and the predicted repayment probability multiplied by the payable amount represents the predicted repayment amount.According to 33 repayment dates,we can get the following results: the RMSE of eachmachine learning algorithm model based on cross-section data is almost the same,both in classification effect and prediction of repayment amount,while the RMSE of LSTM neural network based on time series data is 70 left and right lower than the previous one.It shows that it is feasible to use time series data to build LSTM to study credit situation,and the effect is better than the machine learning algorithm used by previous scholars.Then,considering the cost of all aspects of the platform,this paper provides a more effective cash flow management scheme for the platform,which divides the user repayment into the first ten days,the middle ten days,the last ten days and default,and constructs a double-layer short-term memory neural network(p-b-lstm).It can not only predict the repayment amount of users in each period of time,so as to manage the cash flow of the platform,but also maintain customers,such as reminding users of the repayment date and amount in advance,and listing users in the white list or blacklist.After dividing the repayment date into stages,the following prediction results are obtained: the effect of the phased prediction is more reliable and effective than that of the direct prediction of repayment,the F1 score of xgboost and lightgbm is increased by 46% and 43% respectively,and the RMSE of LSTM is reduced by about 110%.In addition,the RMSE of p-b-lstm is about 50 lower than that of xgboost and lightgbm,which indicates that the long-term and short-term memory neural network has better application value in credit problems.To sum up,the best solution to solve the cash flow management problem of the online loan platform is to use LSTM to forecast the user's repayment date in stages,which can not only effectively predict the repayment situation,but also improve the operation efficiency of the platform,and ultimately help the platform to achieve a virtuous cycle of borrowing and lending and achieve healthy development.