DeepFedKDAD: Deep Federated Knowledge Distillation Based Anomaly Detection in B5G Network Slicing

In order to support the various needs of future applications and different services anticipated in Beyond 5G (B5G) networks, B5G networks will rely on network slicing (NS), a key technology that creates multiple virtual networks on a shared physical infrastructure. As the number of user equipment (UE) connections accessing resources from the B5G NS is expected to increase, security and privacy concerns may arise from abnormal behaviors of these UEs. These security and privacy concerns can disrupt the performance of the B5G NS and impact the slices' ability to effectively provide resources and services to the UEs. Therefore, we propose a novel deep-federated knowledge distillation-based anomaly detection framework, named DeepFedKDAD, to address the se-curity and privacy concerns arising from anomalous UEs. The results show that our proposed anomaly detection framework outperformed baseline studies regarding accuracy, precision, recall and f1-score performance metrics in detecting anoma-lies. Additionally, the results showed that our DeepFedKDAD converges faster as compared to the baseline studies.