Federated Learning-powered DDoS Attack Detection for Securing Cyber Physical Systems in 5G and Beyond Networks

Abstract

Abstract

The advent of 5G networks has revolutionized Safety-Critical Cyber Physical Systems (CPS), such as autonomous transportation, healthcare, and industrial automation, by providing ultra-reliable, low-latency, and high-speed communications essential for real-time decision-making and control. However, these advancements also make 5G infrastructures attractive targets for Distributed Denial of Service (DDoS) attacks, which can severely disrupt network functionality and jeopardize critical services. To address this challenge, we propose a comprehensive approach for securing 5G-enabled CPS through advanced anomaly detection and Federated Learning (FL). Our research introduces an LSTM Autoencoder-based anomaly detection model specifically designed for multivariate time series data from 5G core networks, enhancing the detection of potential intrusions. We leverage FL to collaboratively train and update the Intrusion Detection System (IDS) across decentralized 5G deployments, preserving data privacy and reducing network bandwidth requirements. Recognizing the vulnerability of FL to data poisoning attacks, we also evaluate and implement state-of-the-art defense mechanisms to protect the integrity of the federated model. This research provides valuable insights and recommendations for deploying robust, privacy-preserving IDS solutions in 5G networks, contributing to the advancement of secure and efficient 5G infrastructure for critical applications.