Deep Learning–Guided Mother Optimization Algorithm With Surrogate QoS Modeling for 5G Network Slicing

ABSTRACT The rapid proliferation of heterogeneous 5G services, including enhanced mobile broadband (eMBB), ultra‐reliable low‐latency communication (URLLC), and massive machine‐type communication (mMTC), has significantly increased the complexity of network slice resource allocation. Ensuring efficient resource utilization while satisfying diverse quality‐of‐service (QoS) requirements such as latency, reliability, and throughput remains a challenging task in dynamic network environments. Conventional optimization, reinforcement learning, and deep learning approaches often suffer from high computational overhead or limited capability to jointly consider multiple QoS indicators under rapidly changing traffic conditions. To address these challenges, this paper proposes a surrogate‐assisted deep learning and mother optimization algorithm (DL–MOA) framework for intelligent and QoS‐aware network slice resource allocation. The proposed system first employs a multitask deep learning predictor to classify slice types and estimate initial resource requirements from traffic features. A lightweight surrogate QoS model is then introduced to approximate key QoS metrics, enabling fast evaluation of candidate resource allocations during optimization. Guided by these predictions, the mother optimization algorithm performs multiobjective resource allocation by optimizing bandwidth, latency budget, and reliability margins while satisfying network constraints. Experimental results on 5G traffic datasets demonstrate that the proposed framework achieves 99.9% slice prediction accuracy, improves resource efficiency by 2.6%, reduces SLA violation rates by 24.3%, and lowers decision latency to 265 ms compared with DRL‐based and meta‐heuristic baselines. These results highlight the effectiveness of the proposed approach for real‐time, QoS‐aware network slicing in next‐generation wireless networks.