Federated Learning and Resource-Aware Graph Neural Network for Intrusion Detection in 6G-IoT Driven Healthcare System

Large Language Models (LLMs), enabled by advancements in Transformer architectures, have revolutionized natural language understanding and multimodal processing, finding transformative applications in sectors such as healthcare, intelligent manufacturing, and smart transportation. In healthcare, LLMs facilitate tasks like semantic analysis of medical records, real-time monitoring, and intelligent diagnostics. The emergence of 6G, with its integration of AI, IoT, and edge computing, offers unprecedented opportunities for deploying large models in real-time, privacy-sensitive, and resource-constrained environments. This paper proposes FLARE (Federated Learning and Resource-aware Embedding), a novel framework tailored for IoT-driven healthcare systems. FLARE introduces three technical innovations: dynamic graph modeling with optimal temporal window selection, edge-aware aggregation mechanisms that enhance local feature representation, and adaptive multi-hop embedding strategies that address computational constraints. By adopting federated learning instead of centralized approaches, FLARE enables privacy-preserving collaborative modeling across distributed IoT nodes, while effectively addressing data heterogeneity and reducing privacy risks in network traffic analysis and intrusion detection. Experimental results on CICIDS2017 and UNSW-NB15 datasets demonstrate FLAREs superior performance, achieving F1-scores of 0.935 and 0.927, and AUC-ROC values of 0.947 and 0.938 respectively, outperforming state-of-theart methods while maintaining competitive training efficiency