BFCN: A Novel Classification Method of Encrypted Traffic Based on BERT and CNN

With the speedy advancement of encryption technology and exponential increase in applications, network traffic classification has become an increasingly important research topic. Existing methods for classifying encrypted have certain limitations. For example, traditional approaches such as machine learning rely heavily on feature engineering, deep are susceptible to amount distribution labeled data, pretrained models focus merely global features while ignoring local features. To solve above problem, we propose a BERT-based byte-level convolutional (BFCN) model consisting two novel modules. The first is packet encoder module, which use BERT capture through its attention mechanism; second CNN captures operations. packet-level concatenated traffic’s final representation, can better represent traffic. Our approach achieves state-of-the-art performance publicly available ISCX-VPN dataset service application identification task, achieving F1 scores 99.11% 99.41%, respectively, these tasks. experimental results demonstrate that our method further improves classification.