CAFED-Net: Cross-Adaptive Federated Learning with Dy-namic Adversarial Defense for Real-Time Privacy-Preserving and Threat Detection in Distributed IoT Ecosystems

Abstract

The industrial, urban, and healthcare sectors now require immediate cybersecurity measures due to the rapid expansion of distributed Internet of Things (IoT) networks across these domains. Centralized threat detection systems struggle to adapt to diverse IoT environments that cover large-scale systems with highly sensitive response times. New threats rapidly develop against defense systems that maintain fixed positions, creating both mobile threats and expanding across data domains. Federated Artificial Intelligence (Federated AI) introduces a training methodology that addresses data privacy concerns while safeguarding against severe data breaches that can result from centralization. A new Federated AI system architecture has been created to manage cross-domain threat detection activities in distributed IoT environments through the implementation of dynamic adverse adaptation strategies. The system performs local training operations at distributed nodes and combines them with federated aggregation while implementing an adaptive intelligence-sharing framework. This method establishes resilient mod-els along with domain-independent capabilities that protect data independence. The proposed system utilizes adversarial training as it dynamically adjusts to emerging attack vectors during operation. The detection capabilities, coupled with the adaptability of simulation-based evaluation, demonstrate supe-riority over baseline models during adversarial drift conditions. This research presents an approach that enables real-time privacy-safe IoT threat detection at a scalable level to address evolving cybersecurity threats.