LLM based approaches for traffic prediction in networks traffic

Abstract

In modern communication networks, particularly within the context of 5G and beyond, network slicing has emerged as a key technique to support diverse services with varying Quality of Service (QoS) requirements. Each slice is designed to meet the specific needs of applications such as video streaming, IoT, and ultra-reliable low-latency communications, and must be provisioned with appropriate resources. A major challenge in network slicing is the dynamic and unpredictable nature of network traffic. As traffic is user-generated and varies over time, it cannot be directly controlled by the network operator. This time-varying behavior makes static resource allocation strategies inefficient, potentially leading to congestion, increased delay, or poor resource utilization. Therefore, accurate traffic prediction is essential to enable proactive and adaptive resource management.