Machine Learning-Driven Load Balancing in Edge and 5G Networks: A Comprehensive Survey

Abstract

The proliferation of connected devices and latency-sensitive applications in mobile edge computing (MEC) and fifth-generation (5G) and beyond (B5G) networks has made intelligent load balancing a cornerstone of next-generation distributed systems. Efficient workload distribution across edge servers is essential for minimizing response time and latency, optimizing resource utilization, and reducing energy consumption, yet the dynamic, heterogeneous, and resource-constrained nature of edge environments makes this a persistently difficult problem. This survey provides a comprehensive, load-balancing-centered review of machine learning (ML), reinforcement learning (RL), deep reinforcement learning (DRL), and multi-agent reinforcement learning (MARL) approaches applied across edge, fog, cloudlet, MEC, vehicular, and 5G/B5G environments. It introduces a structured, multidimensional taxonomy that characterizes existing studies across eight orthogonal dimensions, including control architecture, observability, agent design, training strategy, and environmental dynamics, and a unified four-dimensional problem formulation encompassing queuing delay, network delay, energy consumption, and SLA/QoS compliance. To enable normalized and reproducible cross-study comparison, we further define seven complementary evaluation dimensions and an RL-specific taxonomy that explicitly captures algorithmic design choices such as state–action space design, reward shaping, exploration strategy, and safety constraints. The findings confirm that ML-Driven approaches deliver strong adaptability and optimization capability in dynamic environments, while identifying critical open challenges in scalability, real-world validation, sim-to-real transfer, and standardized reporting. This work is intended as a structured reference for researchers, network engineers, and decision-makers developing intelligent load-balancing strategies for edge and next-generation communication systems.