Fortifying 5G network slices using a trust-based secure federated learning framework for attack detection and classification

The rapid evolution of 5G has revolutionized communication by offering high-speed connectivity and supporting various applications. An essential feature of 5G networks is network slices, which enable the creation of multiple virtualized and independent networks on a shared physical infrastructure to provide a dynamic range of services for specific use cases. However, this flexibility also poses significant security challenges, such as Denial of Service (DoS), Distributed Denial of Service (DDoS), Injection, Malware, and Man-in-the-middle (MITM), to network slices. In addition, existing Machine Learning (ML) and Deep Learning (DL) based approaches cannot adapt to network slices’ distributed and dynamic nature, posing privacy threats. Unlike conventional methods, Federated Learning (FL) presents a more advanced alternative with enhanced security and privacy. However, FL aggregation processes remain vulnerable to several attacks, including model poisoning, data poisoning, and Byzantine attacks. Addressing them is essential for unlocking FL’s complete potential. This paper proposes a trust-based client selection technique to secure FL by ensuring that only trusted, non-malicious clients contribute to global model development. In addition, our proposed secure FL framework uses the ResNet-18 Convolutional Neural Network (CNN) to detect and classify attacks in network slices, achieving 97.36% accuracy in non-malicious environments. The proposed approach significantly outperforms in the presence of 60% and 70% malicious clients, and demonstrates 93.35% and 56.38% accuracy, respectively. These results highlight the effectiveness of our secure FL framework for detecting and classifying attacks in network slices, even in the presence of malicious clients. Furthermore, an experimental analysis on the Edge-IIoT dataset demonstrates the generalizability and robustness of the proposed framework.