Graph-Enhanced Hierarchical Multi-Agent Reinforcement Learning for Adaptive Healthcare Coordination in Smart Fog Systems

Fog computing has emerged as a promising solution for resource-constrained, real-time applications, particularly in the healthcare domain. However, efficient task scheduling remains a significant challenge in dynamic environments. This paper introduces the GE-HMARL (Graph-Enhanced Hierarchical Multi-Agent Reinforcement Learning) framework to address task scheduling in healthcare fog computing systems. The proposed framework combines hierarchical reinforcement learning with graph-based context modeling to enhance task allocation, resource management, and real-time decision-making. We evaluate GE-HMARL against traditional scheduling methods, including Random Task Scheduling, Priority-Based Scheduling, Flat RL, and Non-Adaptive Scheduling, using key performance metrics such as task completion time, load balancing efficiency, emergency response time, and energy consumption. The experimental results show that GE-HMARL consistently outperforms the baseline methods, achieving up to 44.7% reduction in task completion time, 28.6% lower energy consumption, and up to 17.6% improvement in load balancing efficiency. Additionally, GE-HMARL achieves the fastest response times, improving by 60% over the next best method. These findings demonstrate the effectiveness of GE-HMARL in optimizing task scheduling for healthcare applications in fog environments, offering a more efficient and scalable solution for real-time, resource-constrained systems.