FEDERATED LEARNING FOR PRIVACY-PRESERVING CYBERSECURITY: AN INTELLIGENT FRAMEWORK FOR REAL-TIME THREAT DETECTION IN IOT NETWORKS

Abstract

The growing Internet of Things (IoT) networks have put new pressures on cybersecurity due to distributed networks, heterogeneous and resource-constrained devices, constant data creation, sensitive data and increasingly complex attacks. The conventional centralized machine-learning methods are still dependent on shifting and consolidating data to central servers, making them susceptible to privacy threats, communication delays, and even critical weaknesses. While Federated Learning (FL) is also considered a paradigm for joint model training without the need to store raw data on the participating devices or edge nodes, it is important to note that FL is not necessarily private or secure, as updates to the models can be vulnerable to adversarial attacks, such as inference, poisoning, and backdoor attacks (Mothukuri et al., 2021; Zhang et al., 2022). In this qualitative study, how FL can facilitate threat detection via privacy-preserving and near-real-time techniques in heterogeneous IoT environments is discussed in detail while an intelligent conceptual framework is designed to combine dimensions that are typically studied separately in the existing literature. The study is conducted using an interpretivist documentary research perspective, in which the text is analyzed using thematic analysis and qualitative content analysis of literature extracted from peer-reviewed publications, authoritative cybersecurity reports, standards and relevant technical documentation. The synthesis highlights common themes in the literature related to privacy preservation, distributed threat detection, communication and computational efficiency, adversarial robustness, model poisoning, differential privacy, secure aggregation, non-IID data, device heterogeneity, explainability, trust, real-time responsiveness, governance and deployment constraints. Previous works show that FL can help to mitigate the reliance on centralized data collection but its utility relies on a delicate trade-off between privacy protection, resource usage, communication overhead, model robustness, interpretability, and operational latency (Kairouz et al., 2021; Mothukuri et al., 2021). Theoretically, the study connects the dots between privacy and distributed intelligence, cybersecurity resilience and trustworthy AI, and practically puts forth a framework to align these dimensions in IoT threat-detection architectures. It stands out from other definitions by considering privacy, security, efficiency, explainability and adaptability in real time as different aspects of the same design challenge. The framework therefore offers a conceptual base for more reliable, privacy respecting, adaptive and decentralised IoT cybersecurity.