Research And Implementation Of Enterprise Financial Risk Prediction System Based On Transfer Learning

In recent years,the Internet has unknowingly affected every aspect of people’s daily life.In the financial field,the achievements related to Internet finance are becoming more and more mature.As more and more enterprises emerge,it is particularly important to predict the risks of enterprises through Internet technology.However,there are still many emerging companies born every year.These companies have less historical data and lack experience in obtaining accurate labels,which increases the difficulty of financial risk prediction and affects the further promotion of inclusive finance.Listed companies have the characteristics of a large amount of data and complete data labels.By learning the historical data of listed companies to predict the risks of such companies,it can better increase financial inclusion.The basic idea of transfer learning is to use the experience of one task to improve the learning of another task,which happens to be able to solve this kind of problems.Therefore,this thesis proposes a series of financial risk characteristic indicators,and proposes a financial risk prediction model based on migration learning,and develops a financial risk prediction system based on this.The main work of this thesis is as follows:(1)According to the classification of enterprise-related data obtained from the Internet,this thesis is divided into financial characteristics,industrial and commercial characteristics,judicial and public security characteristics,recruitment characteristics,intangible asset characteristics,business activity characteristics,equity structure characteristics,public opinion characteristics,and constructs corresponding characteristic indicators.After the construction is completed,this thesis analyzes the importance of the characteristic indicators,and eliminates some indicators with low importance to improve the efficiency of the prediction model.At the same time,the characteristic indicators constructed in this thesis can be used as a reference for future research on financial risk prediction.(2)This thesis proposes an enterprise financial risk prediction model based on transfer learning,which mainly uses the model-based transfer learning method.The model processes source domain data and target domain data through Triplet-loss representation learning and domain adaptation,and then adds a fine-tuning process,in which parameters are gradually frozen for each hidden layer,and then the model is fine-tuned.Then the model output is calculated through an XGBoost classifier to obtain the final risk prediction result.Then,this thesis verifies the reliability of the prediction model proposed in this thesis through ablation experiments and comparative experiments with other models.(3)According to the needs analysis,this thesis designs and develops a financial risk prediction system based on transfer learning.Adopt B/S architecture for development,use JAVA language combined with Springboot framework to realize system back-end functions,front-end use VUE and its related components Element-UI for development work,and use My SQL for data reading and storage in the database.Finally,through the function test and performance test of the system,it is verified that the system designed in this thesis can realize financial risk prediction,and both function and performance meet the expected requirements.