In cellular networks, mobility management and resource allocation are two critical tasks that directly impact the quality of service (QoS) for mobile users. Traditional approaches often address these tasks separately, which may not fully exploit the shared information between them and may be suboptimal in dynamic environments. In this paper, we propose a multi-task learning (MTL) framework that jointly learns handover and resource allocation decisions. Our model uses a shared encoder to extract a compact latent representation from multi-domain features, while two task-specific heads perform predictions. A reconstruction branch is incorporated to regularize the latent representation and improve model robustness. The model is trained and evaluated on a large-scale simulated dataset, generated under comprehensive network conditions and user mobility patterns. The proposed framework achieves competitive performance compared to strong baseline models for both tasks. Experimental results show that the proposed model achieves strong classification performance, reaching 94.15% accuracy for HO prediction and an F1-score of 0.925 for resource allocation strategy classification. More importantly, the joint learning framework provides clear system-level benefits. Compared with SVM, random forest, KNN, and single-task MLP baselines, the proposed method improves average user throughput by up to 15.6% and reduces delay by up to 19.5%, at speed 30 m/s. These results demonstrate that shared representation-based MTL can support joint handover (HO) and resource allocation (RA) strategy decision learning and can improve throughput and delay under the considered simulation setting, while maintaining efficient inference suitable for practical deployment in intelligent RAN systems.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)