Advancing Security in 5G Core Networks Through Unsupervised Federated Time Series Modeling
Saeid Sheikhi, Panos Kostakos
Abstract
The rapid development of fifth-generation (5G) mobile communication technology poses fresh challenges for cybersecurity defense systems. Current intrusion detection mechanisms in 5G networks have shortcomings, particularly in identifying sophisticated cyber attacks. Our study presents a novel approach combining Federated Learning with Long Short-Term Memory (LSTM) networks to enhance cyber threat detection on the GTP protocol within 5G infrastructures. Our approach leverages the collective analytical power of multiple devices to identify cyber threats more effectively. The model validated against two major cyber threats, Distributed Packet Forwarding Control Protocol (PFCP) and IP address spoofing emulated within a specially constructed 5G test environment that mirrors a complex public network infrastructure. The findings demonstrate that our unsupervised FL-LSTM model effectively identifies 5G cyber threats while preserving individual network traffic privacy, highlighting Federated Learning's potential to strengthen 5G and beyond network security.