QoS-Driven Snapshot Policy Modeling for Dynamic Network Slicing
This study presents a comparative machine learning analysis for modeling decision-making processes in dynamic network slicing within 6G networks. Experiments use the 6G Network Slicing QoS Dataset containing network telemetry and Quality of Service (QoS) indicators. Two tasks are considered: (i) prediction of the resource allocation decision and (ii) multi-class classification of the operational network state. Several models, including Random Forest, Gradient Boosting, XGBoost, CatBoost, LightGBM, ExtraTrees, Multi-Layer Perceptron (MLP), and Support Vector Machines (SVM), are evaluated under a unified experimental protocol. Performance is measured using accuracy, precision, recall, and macro-F1 metrics. Results indicate that tree-based ensemble methods achieve strong predictive performance for both resource allocation decisions and QoS-driven network state modeling. While several ensemble learning methods achieved near-perfect performance in the resource allocation task, LightGBM achieved the best results in the network state classification task with 0.99 accuracy and 0.98 macro-F1.