The Design And Implementation Of Financial Risk Control Model Based On LSTM Deep Neural Network

The Chinese government decide to keep push the policy that abolish the limit of secondhand car be sold to other provinces,this should be definitely good news for the growing second-hand car market.It has relatively low barriers as well as quick approval in auto finance area which based on internet.Moreover,it also tries to build a wall based on the credit report,order information and other data in auto finance area.Therefore,it has very important significance for auto finance area to build a risk model that can be iterated quickly.It designs and implements a financial risk control model based on LSTM deep neural network in this paper.The model not only reduce the dependence on financial experts,but also has a rapid iteration capability.This paper has two parts: financial data processing and network design.The aim of financial data processing is solving data missing and data imbalance problem.I used a linear function normalization method to fill data and the Border-line SMOTE algorithm to resamples the data.Finally,I build a deep network based on LSTM.I compared the LSTM model,Logic Regression model as well as XGBoost model based on the financial dataset.The results show that there are about 3% improvement after the data processing.Meanwhile The new LSTM model is better than XGBoost model on AUC(0.8223)and KS(0.5424),it also has a low cost on parameter adjusting and easy to iterate.In summary,the new LSTM model can replace the current XGBoost solution after period of parallel verification in online environment.