Elevating 5G Network Security: A Profound Examination of Federated Learning Aggregation Strategies for Attack Detection
Ioannis Makris, Nikolaos Ntampakis, Θωμάς Λάγκας, Panagiotis I. Radoglou Grammatikis, Sotirios K. Goudos, Vasileios Argyriou, Eleftherios Fountoukidis, Antonio Skármeta, Pablo Fernández Saura, Panagiotis Sarigianndis · 2023
The popularity of 5G networks has resulted in significant advancement and opportunities in connectivity and reliability of communications, but, concurrently, it raised security challenges and privacy concerns due to the distributed and highly dynamic nature of these networks. In particular, while participating devices and nodes in a 5G network need to be resilient against cyber threats, most of them are not allowed to exchange their data, and, therefore, they are limited only to the corresponding patterns identified locally. To tackle this, this paper proposes a federated learning approach to enable different nodes to collaboratively train a unified intrusion detection system while avoiding the direct exchange of data. In our experiments, we tested a number of different federated learning strategies with two (2) base stations that serve as participating clients in a federated learning scheme, while a server orchestrates the training phase. In terms of evaluation, the proposed solution was tested against the 5G-NIDD dataset and produced a high detection rate of 97.89 % accuracy.