A Study On Loan Pricing Model Based On Federaed Learning

With the development of machine learning and artificial intelligence technologies,research on loan prediction in small and medium-sized banks has received increasing attention.Many scholars and researchers in the financial field utilize machine learning and data mining techniques to predict the default risk of small and medium-sized enterprises’ loans.By using loan pricing prediction models,small and medium-sized banks can accurately price loans,thereby better meeting customer needs,improving customer satisfaction,enhancing the bank’s competitiveness,and reducing bank risks,as well as saving manpower and time costs.However,due to the sensitivity of banking data,it cannot be safely circulated.Therefore,current banks are limited to local datasets for training.This results in small and medium-sized banks with limited data samples struggling to independently train precise models,leading to a loss of competitiveness against large banks in the market.Additionally,ensuring the security of customer information during the model training process has become another major challenge for banks.To address the aforementioned issues,this paper proposes a pricing method based on the Federated Learning ensemble algorithm for personalized loan pricing services in small and medium-sized banks.The integration algorithm is incorporated into Federated Learning,resulting in three Federated Learning ensemble algorithms: Stacked-FL,Voted-FL,and Averaged-FL,which adopt the same base learners: Linear,Poisson,and Secure Boost.The Federated Learning ensemble algorithms not only ensure the security of bank customer information but also improve predictive performance.In the experiment,the Lending Club dataset is split to simulate two participating parties with overlapping users and different sample features.The experiment utilizes preprocessing of multi-source heterogeneous data,data sample alignment,encryption training of federated algorithms,and federated ensemble algorithms as the infrastructure.A combination of HASH-256 and RSA encryption and homomorphic encryption algorithms is used to encrypt the alignment and transmission of gradients and losses between participating parties.The experimental evaluation of the Federated Learning ensemble models and base learners is conducted using two evaluation metrics: MAE and RMSE.Based on the experimental results,it can be concluded that the predictive performance of Federated Learning ensemble algorithms is superior to individual models,improving prediction accuracy while ensuring user privacy and security.This can provide better services to customers of small and medium-sized banks.The paper broadens the scope of joint loan pricing in small and medium-sized banks,providing valuable insights for the promotion of personalized loan pricing services.