Privacy-Conscious WiFi Traffic Classification Using non-DPI Features
9th International Balkan Conference on Communications and Networking, Balkancom 2026, Ulcinj, Karadağ, 16 - 19 Haziran 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/balkancom71095.2026.11605012
- Basıldığı Şehir: Ulcinj
- Basıldığı Ülke: Karadağ
- Anahtar Kelimeler: multi-class ML classification, Network traffic classification, network traffic data, WiFi analytics
- Boğaziçi Üniversitesi Adresli: Evet
Özet
We present a machine learning approach to network traffic classification (NTC) problem aimed at improving Internet quality of service. The method preserves user privacy by relying solely on temporal and statistical features collected at the access point, without packet inspection. Traffic is classified into four categories: video streaming, video calls, gaming, and other. To reduce the gap between validation performance and real-world generalization, we introduce a data generation framework called Iterative Data Enhancement and Acquisition (IDEA). Our XG-Boost model trained on a combination of real-world and emulated data achieves an F1macro score of 0.876 on unseen real-world traffic, demonstrating the robustness and effectiveness of traffic classification without packet inspection.