MFTE: Multifactor and fuzzy trust evaluation for federated learning in mobile edge computing

Abstract

Federated learning effectively mitigates privacy leakage issues in mobile edge computing by implementing collaborative training without data sharing, but it also poses challenges to the trust and security of the terminal nodes. There is little research on trust evaluation in federated learning, and existing studies have overlooked the influence of data and reputation trust generated by third-party recommendations. To address these issues, a trust evaluation scheme for federated learning terminal nodes based on multifactor and fuzzy is proposed, incorporating factors such as node behavior, node reliability, and reputation trust. By integrating the current trust and historical trust to obtain the direct trust of terminal nodes and combining it with the reputation trust generated by edge nodes, a comprehensive trust is derived. On this basis, a reputation trust filtering model based on T-S fuzzy logic is proposed to address dishonest recommendations and the uncertainty of reputation trust resulting from malicious attacks. The similarity, timeliness, and external trust of recommendations are analyzed, and fuzzy inference is used to filter dishonest recommendations. The experimental results demonstrate that the proposed scheme can rapidly identify malicious nodes, accurately evaluate node trustworthiness, and effectively filter out dishonest recommendations. Compared with the state-of-the-art scheme, the proposed scheme demonstrates improvements in evaluation accuracy and robustness.