Cross-Silo Horizontal Federated Learning Methods in Network Traffic Analysis

Federated Learning (FL) is a Machine Learning (ML) technique allowing multiple parties to collaboratively train a model without sharing their raw data with each other. This approach is particularly useful in scenarios where data privacy, data sovereignty, and data protection are a concern and ensures organizations comply with data protection laws like the General Data Protection Regulation (GDPR). In network traffic analysis, FL is applicable in scenarios where decentralized data sources are used for training, for example in network traffic prediction, which is a critical task for resource allocation and network optimization. Therefore, in this work, the performance of different aggregation methods in FL for network traffic prediction is examined. Aggregation methods such as weighted federated averaging, secure aggregations, and robust aggregations are considered for a comparative study. The performance of these methods is evaluated on the GEANT project’s public dataset. This work provides insights into the selection of appropriate FL methods in network traffic analysis and sheds light on the trade-offs between accuracy, and robustness in the presence of malicious clients. The results indicate that the combination of median aggregation with Homomorphic Encryption (HE) is a good choice as it provides feasibility for performance, security, and robustness.