Research on Privacy Protection and Secure Sharing Mechanism in Big Data Transaction Circulation Based on Improved Federated Learning Algorithm

Abstract

With the rapid development of the data element market, big data transaction circulation has put forward higher requirements for cross-domain collaborative modeling and privacy protection. Existing methods can reduce the exposure of raw data, but still have shortcomings in non-independent and identically distributed data processing, gradient leakage protection, malicious node identification, and communication efficiency, which limit their applications in complex transaction scenarios. To this end, this paper proposes an improved federated learning (IFL-PSS) secure sharing mechanism oriented to big data transaction circulation. By introducing adaptive weighted aggregation, differential privacy perturbation, and node reputation evaluation strategies, the mechanism realizes privacy enhancement and trusted collaboration during the whole process of model training. Theoretical analysis and simulation verification demonstrate that compared with the traditional FedAvg algorithm, the proposed method can effectively improve model accuracy, reduce communication overhead, and significantly decrease the success rate of member inference attacks, verifying the comprehensive effectiveness of the mechanism in privacy protection, secure sharing, and model performance optimization.