The exponential growth of encrypted network traffic and the increasing adoption of dynamic port allocation techniques have presented significant challenges to traditional traffic classification methods. In response, this paper introduces a novel flow-based online network traffic classification model leveraging machine learning techniques. The proposed approach extracts and processes 29 statistical, temporal, dynamics, and aggregation-based flow features, from 5-second time windows, enabling accurate real-time classification while preserving user privacy. The model is evaluated using network traffic data collected by the authors from various applications, including video streaming, gaming, video conferencing, web browsing, audio streaming, and web streaming. Additionally, the evaluation incorporates six publicly available benchmark datasets to validate the robustness and effectiveness of the proposed model. The obtained results demonstrate superior classification accuracy, consistently exceeding 96% and outperforming existing stateof-the-art approaches in the literature. The proposed method offers a robust and scalable solution for real-time traffic analysis, making it highly suitable for modern network environments and latency requirements.
Rau et al. (Thu,) studied this question.