Boosting Encrypted Traffic Classification Using Feature-Enhanced Recurrent Neural Network With Angle Constraint

With the surge in various types of network traffic and the widespread application of encryption technology, the classification of encrypted traffic plays an increasingly important role in ensuring network security, enhancing quality of service, and managing network traffic. However, most existing methods often suffer from issues such as excessive reliance on manual feature extraction and expert knowledge, unstable classification performance, and lack of transfer learning capabilities. To address these challenges, this paper proposes a high-performance hybrid encrypted traffic classification framework, FERNN-AC. It directly extracts features from raw traffic and fully explores and utilizes the spatiotemporal information of traffic data by integrating specially designed feature enhancement module and temporal feature extraction module in a reasonable manner. It introduces angle constraints and can be combined with meta-learning, thereby improving classification performance while possessing certain transfer learning capabilities. The experiments are conducted on three datasets, and the results shows that compare with relevant baseline methods, FERNN-AC has excellent and stable classification performance.