Spatio-Temporal Beam-Level Traffic Forecasting in 5G Wireless Systems Using Multi-Task Learning

Accurate traffic forecasting is essential for optimizing resource management and enhancing Quality of Service (QoS) in 5G wireless systems. This paper presents a novel multitask learning (MTL) framework for spatio-temporal beam-level traffic forecasting in 5G wireless systems. The approach leverages shared representations across multiple prediction tasks, addressing the challenges of intermittent data patterns, limited time-series lengths, and multivariate complexity. Comprehensive experiments using a high-resolution, beam-level throughput dataset from real-world 5G deployment provided by the AI for Good - International Telecommunication Union (ITU) demonstrate significant performance improvements over conventional machine learning models such as Long Short-Term Memory (LSTM), with the proposed MTL framework consistently achieving superior performance. Specifically, the MSE and MAPE are 0.00428 and 0.4% respectively, using MTL compared to 0.079 and 36.5% using LSTM. These findings highlight the potential of applying MTL for real-world deployment in dynamic network environments in 5G.