From Features to Qubits: A Hybrid Learning Approach for Predicting SDN Link States

Software-Defined Networks (SDNs) significantly enhance network management through centralized control, programmability, and agility. However, accurate link prediction remains challenging due to the complexity and dynamic topology of SDNs. Classical machine learning techniques achieve high accuracy but can struggle to model subtle, nonlinear relationships in rapidly evolving networks. Quantum Machine Learning (QML) techniques exploit quantum phenomena to capture these complex patterns, yet current quantum hardware limitations restrict their standalone applicability. To address these challenges, this paper proposes a hybrid classical-quantum machine learning framework for SDN link prediction. Classical preprocessing, including normalization and feature selection, prepares high-quality input for quantum processing. The selected features are then mapped into quantum states using the ZZFeatureMap, enabling the Quantum Support Vector Machine (QSVM) to detect complex patterns through quantum-enhanced kernels. Rigorous experimentation using synthetic SDN datasets demonstrates that the hybrid QSVM achieves an area under the ROC curve (AUC) of 0.84, an average precision (AP) of 0.74, and an accuracy of 0.84 - closely matching classical SVM and approaching the performance of Random Forest (AUC = 0.86, AP = 0.78, Accuracy = 0.88). The QSVM and classical SVM also demonstrate significantly lower training time ( 0.04s) compared to Random Forest ( 0.29s), highlighting the hybrid model's efficiency. These results provide strong evidence that quantum-enhanced techniques can effectively complement classical models, offering improved robustness and scalability for intelligent network systems.