Message-Passing Neural Network (MPNN) for Large-Scale Healthcare IoT Networks in Smart City

In smart cities, real-time monitoring of patients’ health conditions is crucial for ensuring timely interventions. However, due to the extensive geographical coverage required, manually tracking patients’ conditions poses significant challenges. To address this issue, the Healthcare Internet of Things (H-IoT) networks have been developed to facilitate data transmission. However, given the limited availability of base stations and various signal issues, maintaining optimal wireless performance in H-IoT networks remains a significant challenge. To mitigate these challenges, this article proposes a message-passing neural network (MPNN) algorithm to determine a near-optimal user assignment strategy within the H-IoT network. By leveraging the graphical representation of message passing (MP) and the aggregation-update mechanism of graph neural networks (GNNs), the proposed algorithm effectively combines both the network model and historical data to optimize decision-making. Unlike conventional GNN-based methods, which either treat user assignment as a closed-box regression problem or require centralized training with complex constraints, the proposed algorithm introduces a hybrid architecture that exchanges small-sized messages iteratively and subsequently optimizes via a graph-based rule. The objective function of the proposed algorithm is clearly defined as a nonlinear problem, which is subsequently derived mathematically via the MP rule to find an autonomous decision. The solid mathematical formulation and hybrid algorithm enhance scalability, improve interpretability, and support distributed operations with minimal communication overhead. Consequently, the proposed algorithm achieves near-optimal performance while maintaining low computational complexity. Extensive simulation results demonstrate that the proposed algorithm outperforms conventional methods in terms of different users and cell sizes, adaptability to varying base station power transmission, robustness in diverse wireless path loss environments, and computational complexity.