Recent Advances in Prediction of IoT Traffic Using Gradient Boosting, Auto-Metric Graph Neural Network, and Lyapunov Optimization-Based Predictive Model: A Systematic Review

Abstract

The rapid expansion of the Internet of Things (IoT) has significantly increased network traffic across smart cities, healthcare, industrial automation, and intelligent transportation systems, creating major challenges in traffic management, resource allocation, and Quality of Service (QoS). Traditional statistical and regression-based prediction models are often insufficient for handling the nonlinear, dynamic, and heterogeneous characteristics of IoT traffic data. As a result, advanced machine learning and optimization-driven approaches such as Gradient Boosting, Auto-Metric Graph Neural Networks (GNNs), and Lyapunov Optimization have emerged as effective solutions for intelligent traffic prediction. This systematic review presents a comprehensive analysis of recent advancements in IoT traffic prediction models, emphasizing hybrid frameworks that combine predictive intelligence with optimization capabilities. Gradient Boosting techniques enhance prediction accuracy through ensemble learning and feature interaction analysis, while GNN-based approaches effectively capture spatial and temporal relationships among interconnected IoT devices. Lyapunov optimization contributes to network stability, adaptive resource management, and delay minimization in dynamic environments. The review critically examines methodologies, datasets, evaluation metrics, and performance outcomes reported in recent studies. Furthermore, it identifies important challenges including scalability, real-time adaptability, energy efficiency, security, and data heterogeneity. The study concludes that hybrid intelligent models integrating machine learning with optimization frameworks significantly improve prediction accuracy, latency reduction, and overall IoT network efficiency.