| Tartalom: |
http://hdl.handle.net/10890/64946 |
| Archívum: |
Műegyetem Digitális Archívum |
| Gyűjtemény: |
1. Tudományos közlemények, publikációk
Konferenciák gyűjteményei
Workshop on Intelligent Infocommunication Networks, Systems and Services
4th Workshop on Intelligent Infocommunication Networks, Systems and Services, 2026
|
| Cím: |
Privacy Preserving Network Flow Classification with Federated Learning
|
| Létrehozó: |
Al-Muramadi, Salam Allawi Hussein
Répás, Sándor
|
| Dátum: |
2026-05-27T09:22:31Z
2026
|
| Tartalmi leírás: |
Encrypted transport protocols reduce network visibility while increasing the need for privacy-preserving traffic analysis. This study compares centralized and federated learning for encrypted network event classification, evaluating whether federated learning can maintain competitive performance without sharing raw client data. Experiments are conducted on a QUIC traffic dataset containing encrypted flows from five Google services, using flow-level statistical features. Identical neural network architectures are employed to ensure a fair comparison. Federated learning is evaluated under extreme non-IID conditions, where each client observes traffic from a single service. Results show Federated learning attains approximately 75–76% accuracy under extreme non-IID conditions, with macro-average AUC values exceeding 0.97. These findings indicate that federated learning is a practical and privacy-preserving alternative to centralized approaches for encrypted traffic classification.
|
| Nyelv: |
angol
|
| Típus: |
Könyvfejezet
|
| Formátum: |
application/pdf
|
| Azonosító: |
|
