강건한 양방향 트랜스포머 사전학습 언어모델 기반 암호화 트래픽 분류

Abstract

The proliferation of internet service platforms has led to an increase in the volume and diversity of traffic data. Consequently, the need for traffic classification has become more pressing, necessitating new approaches to encrypted traffic classification. In this paper, we propose the Robust BERT for Encrypted Traffic Classification (RB-ET), a transformer-based model designed to overcome the limitations of traditional DPI methods. The RB-ET model enhances efficiency by removing NSP during pre-training and utilizing Half-Chance Label Prediction (HCLP) to enable learning from unlabeled data as well. Experimental results show that RB-ET has successfully improved the accuracy of encrypted traffic classification and reduced training time compared to existing models.