Multi-Task Learning at The Mobile Edge: An Effective Way to Combine Traffic Classification and Prediction

The rapid growth of mobile data traffic, fueled by a variety of user applications and diverse service needs, has created significant challenges for managing traffic efficiently in next-generation wireless networks. Typically, traditional models for traffic classification and prediction function as separate tasks, which leads to a heavier computational load and less efficient inference. This paper introduces a cohesive solution that employs multi-task learning (MTL) at the mobile edge, allowing for simultaneous traffic classification and prediction in a way that is both resource-efficient and context-aware. The proposed edge-based MTL architecture utilizes shared representations through deep neural networks, facilitating the joint learning of related tasks. This approach not only boosts task performance by leveraging the connections between tasks but also greatly reduces latency and bandwidth usage by processing data closer to its source. By implementing the model on edge servers situated near base stations, we effectively eliminate the need to send data to centralized clouds, achieving real-time intelligence. Comprehensive experiments conducted on real-world mobile traffic datasets show that our model achieves impressive classification accuracy while keeping prediction error rates low. Additionally, when compared to traditional single-task learning models, our edge-based MTL method enhances generalization, shortens training time, and supports adaptive learning in dynamic mobile environments. This research lays the groundwork for a promising framework for intelligent traffic management in 5G and beyond, fostering efficient resource allocation, network slicing, and quality of service (QoS) guarantees across various traffic scenarios.