Artificial Intelligence Techniques for Prediction of IoT Traffic Using Gradient Boosting, Auto-Metric Graph Neural Network, and Lyapunov Optimization-Based Predictive Model: Trends and Challenges

Abstract

The rapid expansion of Internet of Things (IoT) ecosystems has significantly increased network traffic complexity, requiring intelligent prediction mechanisms for efficient resource management. Traditional statistical methods are inadequate in capturing nonlinear, dynamic, and spatio-temporal dependencies present in IoT traffic. As a result, Artificial Intelligence (AI) techniques such as Gradient Boosting, Graph Neural Networks (GNNs), and Lyapunov optimization have gained considerable attention. This paper presents a comprehensive review of AI-driven IoT traffic prediction models, emphasizing emerging trends and challenges. Gradient Boosting techniques provide high accuracy in structured data prediction, while GNN-based models effectively capture spatial relationships in network topologies. Lyapunov optimization offers a robust framework for maintaining system stability and dynamic resource allocation. Recent studies show that hybrid models combining GNN and optimization techniques outperform traditional approaches in terms of throughput, latency, and adaptability. The review analyses recent developments, highlighting advancements in deep learning, federated learning, edge intelligence, and hybrid optimization frameworks. Additionally, key challenges such as scalability, computational complexity, privacy, and real-time deployment are discussed. The study concludes by identifying future research directions, including lightweight AI models, adaptive learning systems, and integration with edge computing for next-generation IoT networks.