Towards Predictive Topology Management in Service-Oriented 6G Networks

An efficient topology management system is essential for enabling dynamic network creation in future 6G networks. This paper presents an enhanced topology change management system designed for a hybrid 6G network, incorporating novel autonomous networking concepts of sub-networking. The main focus of this work is two-fold: i) Topology Detection and Changes Prediction and ii) Topology Adaptation. The former leverages machine learning to anticipate topology changes proactively, whereas the latter optimizes network responses for seamless adaptation to new network conditions. We deploy an ensemble learning approach for prediction that integrates XGBoost, Random Forest, and SVM using a confidence-weighted voting mechanism, and provide their comparative performance analysis. We address three topology change scenarios, as caused by: (1) Handover; (2) Connection Loss; and (3) Sub-network Overload. The results show that XGBoost outperforms other models in detecting handovers and sub-network overloads, whereas the ensemble model refines predictions by reducing uncertainty and improving confidence. SVM remains the most conservative model, minimizing unnecessary topology changes, while XGBoost excels in rapid response scenarios, ensuring fast and confident decisions. Random Forest provides a balanced performance between stability and adaptability. The ensemble model reduces misclassifications, improves adaptation, and enhances topology change accuracy, providing an effective solution for 6G networks.