Research On Pre-loan Credit Rating Method For Small And Micro Enterprises Based On Blockchain And Federated Learnin

Currently,supporting the high-quality development of micro and small enterprises(MSEs)is crucial for the national economic development.However,due to their small scale,limited information,and high risk characteristics,difficulty in financing has always been one of the bottlenecks that hinders their development.Therefore,building a scientific and efficient credit risk assessment system has become a key issue that commercial banks need to urgently research and solve to develop their credit business.In recent years,research on MSEs credit assessment models has made significant progress,but further exploration,research,and indepth study are still needed.On the one hand,there is still a problem of limited data or poor data quality in the financial credit field,with data existing in the form of "islands".On the other hand,protecting users’ data privacy is an important issue that needs to be addressed.To address these issues,this thesis proposes a MSEs credit assessment model based on blockchain and federated learning,with specific solutions as follows:(1)Considering the important reference value of enterprise electricity consumption data and credit data for the operating situation of MSEs,this thesis uses federated learning technology to introduce external data from the power system,the People’s Bank of China credit system,and other sources to supplement the insufficient dimensions of existing data.Multiple perspectives of enterprise information and multidimensional enterprise data are used for risk assessment.Furthermore,based on the secure,transparent,tamper-proof,and decentralized characteristics of blockchain technology,a MSEs credit assessment model architecture(DBFL)based on blockchain and federated learning is proposed.Participating parties can achieve trustworthy data calculation through federated learning and complete trustworthy data storage through blockchain.(2)To overcome the problems of inflexibility and limited application scenarios of singlechain structures in commercial applications,a dual-chain architecture is proposed.The main chain is a traditional public chain responsible for updating the global model.The side chain is a consortium chain responsible for building a trust network.To ensure the quality of the global model,a main chain consensus algorithm based on proof of contribution(Po C)and a side chain consensus algorithm based on proof of reputation(Po R)are proposed to promote the sustainable development of the federated ecosystem.(3)To efficiently identify high-risk customers and reduce bad debt rates,it is necessary to find strong correlated features in heterogeneous data from multiple sources to construct an optimal feature set.Therefore,this thesis introduces a feature selection process and designs a two-step feature selection algorithm(Xm P)based on wrapper and filter methods to provide better interpretability and accuracy.This thesis conducted empirical analysis using real business data from a commercial bank,and the results show that the proposed approach has more efficient data processing and secure data protection capabilities while achieving the required accuracy.It can address the problems of information asymmetry and data isolation in MSEs credit risk assessment,demonstrating the practical value of the theoretical model proposed in this thesis.