HC-MVGAE: An Encrypted Traffic Classification Method Based on Contrastive Learning

With the wide application and continuous evolution of traffic encryption technology, improving the accuracy of encrypted traffic classification has become a key technical challenge for ensuring network security and effective management. Existing methods leverage deep learning for feature extraction but overlook common features among similar samples, which lead to a performance bottleneck in classification accuracy. In this paper, we propose a multimodal graph self-encoder model (HC-MVGAE) based on hierarchical contrast learning. The model utilizes contrast learning to compare samples at both the flow level and packet level, thereby enabling the learning of common features among them. Through training, it improves the classification accuracy and robustness of the model while reducing computational cost. Experimental results show that the proposed contrastive learning-based encrypted traffic classification model outperforms multiple baseline models on the ISCX VPN-nonVPN and ISCX Tor-nonTor datasets in four metrics: accuracy, precision, recall, and F1-score.