Meta-Context Graph Intelligence for Predictive Task Offloading In Federated Edge-Cloud Iot Systems

Abstract

The rapid growth of heterogeneous Internet of Things (IoT) infrastructures and multi-tier edge–cloud ecosystems has intensified the demand for intelligent, context-aware task offloading mechanisms capable of handling dynamic workloads, unpredictable connectivity, and resource variations. Traditional static or heuristic offloading frameworks often suffer from delayed decision-making and suboptimal resource utilization, resulting in increased latency and energy consumption. To address these challenges, this paper introduces Meta-Context Graph Intelligence for Predictive Task Offloading in Federated Edge-Cloud IoT Systems, a novel framework that employs federated meta-learning integrated with graph neural network (GNN) topology modeling to enable proactive resource allocation. The approach constructs a multi-layer heterogeneous graph representing network topology, device context, energy states, mobility patterns, and workload variations, enabling adaptive and future-aware offloading decisions. A predictive meta-context encoder refines graph features collaboratively across distributed edge nodes using federated training without sharing raw data, ensuring privacy preservation and scalability. Experimental analysis demonstrates improved latency reduction, enhanced task completion success rate, and reduced energy consumption when compared to recent task offloading strategies. The proposed framework offers an intelligent, privacy-preserving, and adaptive decision model suitable for next-generation industrial IoT and 6G infrastructures.