Distributed detection of DDoS attack on 5G network slices using federated learning

The advent of 5G networks has revolutionized telecommunications by enabling flexible and efficient service deployment through network slicing. These slices ensure logical isolation among various services, facilitating tailored Quality of Service (QoS) and security policies. However, while network slicing enhances service differentiation, it also increases architectural complexity and broadens the attack surface, making slices more susceptible to cyber threats specifically Distributed Denial-of-Service (DDoS) attacks. Furthermore, the rapid proliferation of connected devices in 5G networks, driven by diverse service offerings, exacerbates security risks. Due to the growing attack landscape, Machine Learning (ML) based Intrusion Detection Systems (IDS) are getting popular over traditional IDS. Since network slices are used by different verticals, they often have strict privacy requirements. Conventional ML-based IDS are centralized in nature, combining all data on a central server, which fails to uphold slice-level privacy constraints This thesis presents FLAccShield, a Federated Learning (FL)-based privacy preserving IDS framework for DDoS detection in 5G networks through collaboration among multiple slices. This distributed approach enables model training across slices while ensuring that data remains local, thereby preserving privacy and minimizing network overhead. The framework is rigorously evaluated for supervised and unsupervised settings for both Independent and Identically Distributed (IID) and non- IID datasets against a varied range of features against three baseline FL strategies: FedAvg, FedProx, and QFedAvg. We also compared FLAccShield’s performance against slice-wise Independent Local Training. The experimental results demonstrate that FLAccShield achieves F1-scores up to 32% higher than traditional aggregation method for FL such as FedAvg and 5% higher than Local Training in supervised settings with IID data. In unsupervised settings, it outperforms state-of-the-art FL methods such as FedAvg, FedProx, and QFedAvg by 0.4%, 1%, and 32%, respectively. Under non-IID conditions, FLAccShield improves by 3% over FedAvg and FedProx, and 2% over QFedAvg, respectively. These findings highlight FLAccShield’s robustness, efficiency, and its ability to preserve privacy in complex 5G environments.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.