Federated Continual Meta-Learning and Graph Neural Network-Based Predictive Self-Healing Framework for Privacy-Preserving Hybrid Iot Communications

Abstract

The rapid expansion of heterogeneous Internet of Things (IoT) applications has created a demand for communication frameworks that are simultaneously private, adaptive, and resilient to failure. Centralized optimization schemes struggle to meet these requirements, as they raise privacy concerns, create single points of failure, and react to faults only after service degradation has already occurred. To address these limitations, this paper proposes a Federated Continual Meta-Learning and Graph Neural Network-based Predictive Self-Healing (FCML-GNN) framework for privacypreserving hybrid IoT communications. The framework combines federated learning for decentralized model training without raw-data sharing, a continual meta-learning mechanism for rapid, forgetting-resistant adaptation to unseen network conditions, and a graph neural network that models dynamic topologies to proactively predict link failures, congestion, and energy depletion. These predictions drive a self-healing controller that performs adaptive routing, protocol switching, and resource reallocation before faults fully manifest. Extensive simulation results across networks of 50-400 nodes show that FCML-GNN consistently outperforms federated-averaging, deep-reinforcement-learning, and graphonly baselines in packet delivery ratio, fault-recovery time, energy efficiency, and overall resilience, offering a scalable and privacy-aware communication intelligence solution for smartcity, industrial, healthcare, transportation, and smart-grid IoT deployments.