Research On Internet Cash Loan Forecast Based On XGBoost Algorithm

In recent years,due to the rapid development of artificial intelligence and big data,the Internet cash loan industry has also opened the curtain.Once released,internet cash loan is popular with the public because of its fast approval,fast lending characteristics.To some extent,Internet cash loans fill the gap in the credit sector,enabling domestic sub-borrowers to enjoy financial services and helping to develop inclusive financial policies.However,there is also a high risk behind the convenience and facility,the most prominent of which is the problem of high default and high interest rates.There is an interaction between the two,high interest rates can promote the probability of default through the role of reverse selection,similar to the expulsion of good currency.Further analysis,the high interest rate reflects the high cost of funds of the Internet cash loan platform,and the high cost of capital is due to the Internet cash loan platform can not match the customer's loan needs and the company set the amount of borrowing caused by the low efficiency of the use of funds.To sum up,one way to reduce the risk of high interest rates on Internet cash loans is to set the amount of borrowing per user.This paper proposes a new method of forecasting the amount of Internet cash loan,which is different from the quota forecasting method which divides the quota interval,which is more accurate in fitting the user's borrowing needs and making the allocation of the funds of the cash loan company more reasonable.Based on the data set of Jing Dong Financial Credit Demand Forecasting Contest as empirical data,after decrypting,cleaning up,exploring analysis and feature engineering,the XGBoost machine learning algorithm with high precision and small variance is used to establish the regression model,predict the customer's borrowing needs for the next month,and set a matching borrowing quota according to the demand.The empirical results show that the model used in this paper can predict the absolute error ratio of at least 70% of the customer's credit demand within 50%,the absolute error ratio of the forecast of at least 30% of the customer's credit demand is within 20%,and the average error is about 230?,which shows that the model has some practical significance.Using this algorithm to predict the credit line,compared with the traditional method of generalizing the customer quota,can more accurately quantify customer demand,improve the user's experience,improve the efficiency of the use of funds.Subsequently,a further analysis of the importance of thecharacteristics can be seen,the user's total historical borrowing on next month's borrowing amount has the greatest impact factors.At the same time,some popular lying that factors that may have a greater impact on demand are less than they might be in the evidence,such as gender and consumption.