Mitigating DDOS Attacks in Virtualized Network Functions (VNF) Using Federated Learning: A Privacy-Preserving Collaborative Defense

Abstract

The rapid integration of Network Function Virtualization (NFV) in 5G-Advanced and 6G infrastructures has introduced significant security vulnerabilities, most notably in Distributed Denial-of-Service (DDoS) attacks. Modern DDoS threats have evolved into sophisticated multi- vector campaigns that leverage AI to bypass traditional static defenses. This paper proposes a novel framework for DDoS mitigation using Federated Learning (FL). Unlike centralized AI models that require the transfer of sensitive network traffic logs to a central server—raising privacy and latency concerns—the proposed FL-VNF architecture enables distributed VNF nodes to train local detection models. Only the model parameters (weights) are shared and aggregated using the FedAvg algorithm. This research evaluates the framework's efficacy against the CICDDoS2019 dataset, demonstrating a detection accuracy of 99.2% with significantly reduced communication overhead and enhanced data privacy. The study further provides an extensive review of 25 seminal and recent works to contextualize the findings within the current technological landscape.