Towards Reusable Models in Traffic Classification

The machine learning communities, such as those around computer vision or natural language processing, have developed numerous supportive tools. In contrast, the network traffic classification field falls behind, and the lack of standard datasets and model architectures holds the entire field back. This paper aims to address this issue. We introduce CESNET Models, a package comprising pre-trained deep learning models tailored for traffic classification. The included models are trained on public datasets for the task of web service classification. Using the new package, researchers and practitioners can skip model design from scratch and the collection of large datasets but instead focus on fine-tuning and adapting the models to their specific needs, thus accelerating the pace of research and development in network traffic classification.