AN AI-ENABLED FEDERATED DEEP LEARNING CYBERSECURITY FRAMEWORK FOR DECENTRALIZED IoT ECOSYSTEMS: A PRIVACY-PRESERVING APPROACH TO REAL-TIME THREAT DETECTION AND ADAPTIVE DEFENCE

Abstract

The rapid growth of Internet of Things (IoT) systems has brought profound cyber security issues because of their decentralized topology, device heterogeneity, and diminished processing resources. Centralized security models in conventional methods are less appropriate, especially where data privacy, real-time performance, and network robustness are essential. This paper proposes an innovative AI-based Federated Deep Learning (FDL) cyber security system to identify and respond to cyber attacks in distributed IoT settings without compromising data privacy. The suggested model combines Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks in a federated learning framework, aided by differential privacy and homomorphic encryption for secure online model training on the edge devices. The system was tested and evaluated with a live network trace-based IoT testbed as well as publicly available datasets (TON_IoT and IoT-23). Experimental results show the model detected 96.2% accuracy, 31.6 milliseconds average latency, and privacy leakage was minimized to 7.4% under membership inference attacks. The adaptive defence mechanism allowed the model to effectively counter novel, zero-day attacks with its detection capability improved from 85.1% to 91.8% upon five rounds of updating. These results prove the feasibility and effectiveness of federated AI architectures for real-time, privacy-enhancing cybersecurity in decentralized IoT systems.