Blockchain-Assisted Federated Explainable Deep Learning Framework for Real-Time Cyberattack Detection in IoT-Enabled Healthcare Systems

Abstract

The extensive use of IoT gadgets in clinical facilities has expanded the attack surface for attackers and thus increased the risks associated with patient safety and privacy of medical data. In order to counter the rising threat, we propose an IoT-based healthcare cybersecurity architecture called Blockchain-Assisted Federated Explainable Deep Learning (BFEDL) which is designed for real-time cyber-attack detection. The BFEDL framework includes three key components: (i) a federated learning approach which maintains patient data on the device level through local training of detection models at distributed IoT edge nodes; (ii) a hybrid deep learning approach which uses the combination of CNN and BiLSTM networks in order to recognize both traffic patterns and temporal relationships in network flows; and (iii) a blockchain-controlled trust layer where the model update process integrity and cryptographic auditing are ensured by Hyperledger Fabric smart contracts, which secure updates from being tampered with during the federated aggregation process. In order to comply with healthcare artificial intelligence regulations, the BFEDL framework incorporates explainable AI techniques SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) producing per-alert explanations understandable by clinical informatics specialists without knowledge of machine learning. Evaluating our framework on the CICIOT2023, N-BaIoT, and TON_IoT datasets, we achieved detection accuracies of 99.41%, 99.18%, and 98.97%, respectively, while having macro-averaged F1-scores exceeding 99.1% for each dataset. Apart from its detection capabilities, BFEDL framework reduces the amount of inter-node communications by 34.2% compared to traditional central learning approaches and requires 1.83 seconds for blockchain verification, thus ensuring compatibility with real-time detection windows in clinical practice.