High-Efficient and Few-Shot Adaptive Encrypted Traffic Classification With Deep Tree


Qiang Wang, Wenhao Li, Huaifeng Bao, Zixian Tang, Wen Wang, Feng Liu, Lingyun Ying.


Although network traffic classification has been investigated for decades, the core challenges, including the complex and capricious conditions of network traffic, and the practical application of models, remain unsolved. Meanwhile, the extensive usage of encryption protocols makes encrypted traffic classification become a new challenge. The rapid iteration of network traffic brings the scale drift of encrypted traffic classification. While bulky deep-learning-based methods can barely satisfy the lightweight demand in real-world scenarios.

To solve this, we propose a efficient encrypted traffic classification method using Deep-Tree with multi-grained scanning and cascade tree to perform high-speed learning and multi classification task. It has the classification accuracy and representation ability of depth model with lightweight computing expenses. The self-adaption and expandable ability of the model make it suit different traffic scenarios without specific model adaptation. The experimental results show that the proposed method achieves superior performance compared with state-of-the-art methods. Particularly, our method can dynamically adapt traffic classification tasks at different scales. The accuracy of the our method in the few-shot evaluation is on average 31.34% higher than the averages F1 score of the baseline methods.