Hierarchical Machine Learning for Regional Resource Allocation in Cloud-Edge Collaborative Environments

Abstract

This study aims to develop a hierarchical machine learning approach based on a cloud-edge collaborative architecture to improve the real-time performance, efficiency, and fairness of regional resource allocation. First, a cloud-edge framework is designed to operate under heterogeneous multi-regional sensing conditions. A task partitioning and data synchronization mechanism is constructed to support dynamic collaboration. Second, a multi-algorithm fusion strategy is employed at the model level. On the edge side, lightweight reinforcement learning and adaptive clustering algorithms enable rapid local policy perception and response. On the cloud side, graph neural network and transfer learning model are deployed to optimize global resource scheduling. Third, a regional resource sensing system is built, incorporating multidimensional indicators such as economy, energy, and employment. A rule engine and feedback mechanism are integrated to enable dynamic closed-loop scheduling. The experimental platform uses real Internet of Things (IoT) data and simulated environments, covering nine representative regions led by manufacturing, services, and agriculture. Testing is conducted across first-tier cities, second-tier cities, and rural areas. Results show that the proposed method achieves an average resource fulfillment rate of 89.6% ± 1.0%, outperforming traditional rule-based methods by 7.5% and centralized models by 3.3%. The average scheduling delay is maintained within 1.6 ± 0.1 seconds, and system resource utilization reaches 74.6% ± 1.1%. In abnormal scenarios, such as edge node failures or cloud service interruptions, the system maintains a task completion rate above 88%, demonstrating strong robustness. Compared with baseline models, resource redundancy in highly dynamic environments is reduced to 16.5% ± 0.8%. The study demonstrates that the proposed hierarchical machine learning approach based on cloud-edge collaboration can achieve efficient resource allocation in complex multi-regional settings, showing strong practical deployment value and scalability potential.