Machine Learning-Enabled Edge Intelligence for IoT Communication Systems: A Structured Review

Abstract

This structured review investigates how machine learning-enabled edge intelligence can improve the performance, efficiency, and resilience of Internet of Things communication systems under constraints of latency, bandwidth, energy, privacy, and device heterogeneity. A structured literature review was conducted using peer-reviewed studies from major academic databases, with the selected work classified according to learning paradigms, edge deployment strategies, communication functions, application domains, evaluation metrics, and practical limitations. The synthesis shows that supervised learning, deep learning, reinforcement learning, federated learning, lightweight model compression, and Tiny Machine Learning can support adaptive scheduling, intelligent routing, computation offloading, bandwidth allocation, anomaly detection, and privacy-preserving collaboration across device-edge-cloud architectures. The reviewed evidence indicates that these approaches can reduce end-to-end delay, communication traffic, and energy consumption while improving resource utilization, local autonomy, and responsiveness in smart cities, industrial systems, healthcare, agriculture, and intelligent transportation. However, performance gains remain strongly dependent on communication conditions, model size, data distribution, hardware capability, synchronization overhead, and security requirements. The review further identifies unresolved challenges involving non-independent and identically distributed data, unstable wireless links, model adaptability, privacy leakage, adversarial threats, and the absence of standardized multi-objective benchmarks. It concludes that future Internet of Things systems should adopt communication-computation-learning co-design, lightweight and adaptive models, privacy-aware distributed intelligence, and cross-layer orchestration to achieve scalable, trustworthy, and energy-efficient edge intelligence.