AdaFed-HybridGAN: Adaptive Federated Aggregation with GAN-Driven Hybrid Model–Data Synthesis for Efficient Cyber Attack Detection in Edge Computing

Open Webpage

Zeseya Sharmin, Yong Xiang, Md Palash Uddin, Feifei Chen

Published: 11 Feb 2026, Last Modified: 05 May 2026CrossrefEveryoneRevisionsCC BY-SA 4.0
Abstract: Federated Learning (FL) has emerged as a promising paradigm for privacy-preserving cyber attack detection in edge computing environments. However, conventional FL approaches face significant challenges. They suffer from communication inefficiency and slow convergence when operating over non-Independent and non-Identically Distributed (non-IID) data across heterogeneous edge servers. To address these limitations, we propose AdaFed-HybridGAN, a novel adaptive federated aggregation framework that integrates model parameter aggregation with selective sharing of synthetic attack patterns generated by a temporal sequential recurrent network-enabled generative adversarial network. Unlike traditional FL methods, AdaFed-HybridGAN shares more than just model updates. It disseminates high-quality synthetic attack samples to augment local datasets while preserving privacy. A quality-aware selection mechanism is introduced to identify synthetic samples with low sequential synthesis data loss. This ensures that only realistic and informative attack patterns are transmitted. Additionally, we design an adaptive aggregation strategy that dynamically adjusts client contribution weights based on both model performance and synthetic data quality. Our theoretical analysis shows that the proposed hybrid aggregation approach achieves tighter convergence bounds than model-only aggregation, particularly under non-IID conditions. Extensive experiments on the ToN_IoT and CSE_CIC_IDS_2018 datasets validate our approach. AdaFed-HybridGAN improves accuracy by up to 6.8%, accelerates convergence by 34%, and reduces communication rounds by 42% compared to state-of-the-art methods. These improvements are achieved, including FedAvg, FedProx, FedGAN-IDS, and FedTSRGNet, while maintaining strong privacy guarantees.