We are pleased to announce that our paper titled “Encrypted Traffic Classification Using 1D, 2D, and Hybrid CNN Architectures” has been accepted for publication at the 18th International Conference on Security of Information and Networks (SINCONF 2026).
The study, conducted by Can Sarper Döğdü and Tacha Serif, investigates deep learning approaches for classifying encrypted network traffic without decrypting its payload. The research addresses the growing need for accurate and efficient traffic classification in modern networks, particularly in the presence of encryption, dynamic ports, and VPN tunneling.
The paper presents a systematic comparison of 1D CNN, 2D CNN, and hybrid CNN architectures using the ISCX VPN/NonVPN dataset. The evaluated approaches transform encrypted traffic into different representations, including sequential byte streams, bidirectional-flow images, grayscale image representations, and RGB representations, allowing both sequential and spatial patterns to be explored.
Experimental results demonstrate that the proposed deep learning approaches can achieve high classification accuracy while maintaining computational efficiency. The 1D CNN architecture achieved the highest overall classification performance, reaching 93.7% accuracy, while the 2D CNN model provided the lowest computational cost. The study also evaluates hybrid approaches that combine sequential and image-based representations, offering a comprehensive comparison of their accuracy, inference speed, memory usage, and suitability for real-time deployment.
This work demonstrates the potential of deep learning to improve encrypted traffic classification without relying on payload decryption. The findings provide practical insights for applications in network security, intrusion detection, traffic analysis, policy enforcement, and VPN traffic identification, while highlighting important trade-offs between classification performance and computational efficiency.