| Summary: | Aiming at the problems of traditional encrypted traffic classification methods limited by the imbalance of dataset classes and the unreliability of the features used in complex network environments, an encrypted traffic classification method based on parallel traffic graph and graph neural network was proposed. Firstly, the traffic graphs were constructed from the packet header and payload perspectives to emphasize their differences. Then, an improved graph attention network was introduced to extract effective information from the parallel traffic graphs. Next, a feature cross-fusion attention module was used to fuse the extracted information, achieving a more robust feature representation. Finally, classification was performed using fully connected layers and a Softmax layer. Experiments show that the proposed method achieves better results on the ISCX-VPN, ISCX-nonVPN, ISCX-Tor, and ISCX-nonTor datasets, with accuracies of 96.88%, 90.62%, 99.24%, and 98.13%, respectively, significantly enhancing encrypted traffic classification performance.
|
|---|
