Design And Implementation Of Loan Overdue Prediction System Based On LSTM-CNN

During the past few years,banks are faced with higher requirements of risk control with the continuous improvement of residents' income,the gradual enhancement of consumer awareness,and the increase of bank loan business.At present,the internal transaction information of bank customers is increasing rapidly,which makes data management more difficult.In addition,the evaluation method of post loan customers is still more traditional,the accuracy of evaluation method is low,the business cycle is long,and the bank needs a more efficient and accurate post loan evaluation model.A personal credit scoring method based on long short term memory and convolutional neural network is proposed in this thesis,which can more accurately predict the possibility of overdue loans and give early warning in time.The data of this thesis is from the internal data of a rural commercial bank in China.This thesis selects the evaluation indexes suitable for loan overdue prediction model,effectively balances the data and selects the characteristics.The behavior data of each user is encoded into a matrix containing time dimension and behavior dimension.This thesis establishes the prediction model of scoring card and the model of loan overdue prediction.The neural network is used to predict the score of the scorecard in the model of the score card prediction,which reduces the human intervention.LSTM-CNN model is used to forecast the overdue loan,which improves the accuracy of the overdue prediction results.The loan overdue prediction system in this thesis mainly includes three functional modules,which are score card module,prediction module and web service module.The score card module is the basic module,which is responsible for data processing and score prediction of scoring card.The prediction module is the core function module,which realizes the core prediction function of the system.The web service module is visualization module,which realizes the visualization display of customer portrait label and the management and analysis of customer data,and completes the interaction between the application layer and the business layer.This thesis makes an experiment on loan overdue prediction based on real post loan data and the experimental results show that LSTM-CNN model is better than single LSTM model and CNN model.The AUC value of the LSTM-CNN model is 0.85,KS value is 0.62,accuracy is 0.86,accuracy reaches 85%.The recall of the LSTM-CNN model is 0.76,which is higher than the 0.66 of the CNN model and the 0.56 of the LSTM model respectively,which is a big improvement.