A novel and effective encrypted traffic classification method based on channel attention and deformable convolution

Abstract

With the rapid development of the Internet and 5G communication technologies, encrypted traffic classification, recognized as a fundamental task in network management, has gained increasing attention from researchers. An accurate traffic classification is crucial for improving network service quality, optimizing network operations, and ensuring information security. However, most of the existing neural network-based methods lack attention to key information and have insufficient traffic feature extraction capability. Motivated by the above limitations, we proposed a novel and effective end-to-end model based on channel attention and deformable convolution, called CAD-Net. Specifically, a novel channel attention mechanism and a powerful deformable convolution technique are introduced in the residual network, which provides the model with the capability of focusing on the most relevant features and capturing the implicit features among the traffic bytes. Furthermore, our class-aware weighted cross-entropy loss function significantly improves the performance of CAD-Net under class-imbalanced conditions. We comprehensively evaluated the effectiveness of CAD-Net on two public datasets, and the experimental results demonstrate that our model achieved the best performance compared with existing methods on both encrypted traffic service and application classification tasks.