FL-TAB-DEN: Federated Learning-Enhanced Trust-Aware Blockchain Framework with Adaptive Intrusion Detection for Secure IoT Healthcare Systems

Abstract

IoT-based healthcare systems continue to face evolving cyber threats including Sybil attacks, model poisoning, and advanced persistent intrusions that static security frameworks cannot adapt to over time. The TAB-DEN framework — Trust-Aware Blockchain with Deception-Based Edge Nodes — established a solid quantum-resistant, blockchain-anchored security baseline but relied on a fixed intrusion detection model. This paper presents FL-TAB-DEN, an extension that integrates privacy-preserving Federated Learning (FL) into TAB-DEN’s hierarchical edge architecture to enable adaptive, continuous, and distributed intrusion detection. FL-TAB-DEN trains a CNN-LSTM hybrid intrusion detection model across Secondary Edge Servers (SES) without centralizing raw healthcare data, aggregating model updates via a differentially private FedAvg protocol. The Deception-Based Edge Nodes (DBENs) are augmented with an FL feedback loop, enabling honeypot profiles to dynamically evolve with emerging attack patterns. Trust scores are recalibrated in real time using federated model confidence outputs. All model updates and intrusion events are logged immutably via the Trust-Aware Blockchain-based Signature Scheme (TABS) using CRYSTALS-Dilithium lattice-based post-quantum signatures. Simulation results demonstrate that FL-TAB-DEN achieves a detection accuracy of 97.2%, a 42% reduction in false positive rate and approximately 26–27% reduction in computational overhead versus TAB-DEN and centralized federated baselines, while maintaining sub-second authentication latency across datasets of up to 50,000 records. These results establish FL-TAB-DEN as a scalable, adaptive, and quantum-resilient security framework for next-generation IoT healthcare infrastructures.