FDGM: A Lightweight Model for Privacy-Preserving Encrypted Traffic Classification

As encryption technologies like SSL/TLS, VPN, and Tor grow more widespread, encrypted traffic classification (ETC) is essential for network management and security. Traditional methods struggle with encrypted data and dynamic ports, posing accuracy and privacy challenges. This paper introduces FDGM (Flexible Deep Grouped Dilated Model), a lightweight and privacy-preserving model for encrypted traffic classification. FDGM efficiently captures spatial and temporal features from traffic packets, achieving 99.3% accuracy on the ISCX-VPN dataset and 99.8% on ISCX-Tor. It outperforms models like Fasttraffic and ET-BERT in both accuracy and computational efficiency.