Our latest research paper on encrypted network traffic classification using hybrid deep learning architectures has been accepted for publication at the 14th International Conference on Future Internet of Things and Cloud (FiCloud 2026), which will take place in Granada, Spain.
We are pleased to announce that our paper titled “AttentionNet: Enhancing Encrypted Traffic Classification Accuracy via Attention Mechanisms” has been accepted for publication at FiCloud 2026.
The study, conducted by Kıvanç Onat Türker and Tacha Serif from the PerSystLab of Yeditepe University, introduces AttentionNet, a hybrid deep learning framework designed for encrypted network traffic classification. The proposed approach treats encrypted traffic analysis as a computer vision problem by transforming raw packet flow data into grayscale image representations.
The architecture combines Convolutional Neural Networks (CNNs) with Transformer-based attention mechanisms to capture both local spatial features and long-range dependencies within encrypted traffic flows. This hybrid design enables the system to classify encrypted and VPN-tunneled traffic without relying on payload decryption or manual feature engineering.
The model was evaluated using the ISCX VPN-NonVPN benchmark dataset across 12 application classes. Experimental results demonstrate that AttentionNet achieves an overall classification accuracy of 86.45% with a Macro F1-Score of 0.88, showing strong performance in distinguishing both VPN and non-VPN traffic categories.
This work contributes to ongoing research in privacy-preserving network intelligence and demonstrates the potential of combining computer vision techniques with attention-based deep learning architectures for modern cybersecurity and encrypted traffic analysis applications.