An Adaptive Edge-Cloud Computing Architecture for Real-Time Intelligent Resource Allocation Using Machine Learning

Abstract

The convergence of edge and cloud computing has created a flexible computational continuum for supporting latency-sensitive, data-intensive, and intelligent applications. However, dynamic workloads, heterogeneous resources, mobility, and changing network conditions make real-time resource allocation increasingly complex. This review examines adaptive edge–cloud computing architectures for intelligent resource allocation using machine learning (ML), emphasizing architectural design, resource management, learning strategies, performance optimization, applications, and emerging research directions. Supervised, unsupervised, deep, reinforcement, multi-agent, federated, and graph-based learning approaches are discussed in relation to workload prediction, task offloading, resource provisioning, service migration, and distributed decision-making. The review highlights the importance of closed-loop adaptation, where resource states are continuously monitored, future demands are predicted, allocation decisions are executed, and resulting performance is evaluated for subsequent policy refinement. Key optimization objectives include latency, energy efficiency, resource utilization, throughput, cost, quality of service, and reliability. Applications across IoT, industrial systems, smart cities, connected vehicles, healthcare, and next-generation networks demonstrate the broad relevance of adaptive resource intelligence. Persistent challenges include heterogeneity, scalability, communication overhead, security, privacy, model drift, and generalization. Future research should prioritize continual learning, explainable intelligence, digital twins, green computing, AI-native 6G, and autonomous orchestration for scalable, secure, sustainable, and resilient real-time edge–cloud computing environments worldwide.