Federated Learning Architectures For Privacy-Preserving Intrusion Detection In IoT Networks

The rapid proliferation of Internet of Things (IoT) devices and smart infrastructure has led to an exponential surge in network traffic, rendering traditional security perimeters increasingly vulnerable. Intrusion Detection Systems (IDS) serve as a critical frontline defense; however, conventional centralized Machine Learning (ML) models are struggling to reconcile high-volume data processing with stringent privacy regulations such as the General Data Protection Regulation (GDPR). Federated Learning (FL) has emerged as a pivotal decentralized paradigm, allowing edge devices and organizations to cooperatively train global models while keeping raw data local, thereby ensuring privacy and reducing model-offloading bandwidth consumption. This review provides a comprehensive analysis of the evolution of FL-based IDS, focusing on its implementation within Industrial Control Systems (ICS) and smart manufacturing environments. We systematically examine the primary technical hurdles facing these architectures, specifically focusing on statistical heterogeneity (non-IID data), communication overhead in resource-constrained networks, and vulnerability to adversarial machine learning attacks such as poisoning and evasion, Furthermore, we discuss specialized integrations with Information-Centric Networking (ICN) and the efficacy of deep learning architectures, such as Long Short-Term Memory (LSTM), in enhancing detection accuracy. The paper concludes by identifying future research avenues, including the need for enhanced model interpretability and robustness against adaptive adversarial threats