CESNET-TLS-Year22: A year-spanning TLS network traffic dataset from backbone lines

Area of Science:

• Computer Science
• Network Security
• Machine Learning

Background:

• Network traffic classification (TC) is vital for network operations and security.
• Machine learning (ML) models are increasingly used for TC due to their ability to identify complex traffic patterns.
• A significant challenge in developing effective ML-based TC is the scarcity of representative real-world datasets.

Purpose of the Study:

• To address the lack of comprehensive datasets for network traffic classification.
• To introduce the CESNET-TLS-Year22 dataset, capturing a full year of TLS traffic evolution.
• To facilitate the evaluation and enhance the robustness of TC models in production network conditions.

Main Methods:

• Collection of Transport Layer Security (TLS) traffic data over a one-year period from an Internet Service Provider (ISP) network.
• Inclusion of 180 distinct web service labels within the dataset.
• Extraction of standard TC features, including packet sequences, for analysis.

Main Results:

• The CESNET-TLS-Year22 dataset provides a unique, year-long perspective on TLS traffic dynamics.
• The dataset enables detailed assessment of TC model performance over time.
• It allows for the evaluation of TC model robustness against evolving network conditions.

Conclusions:

• The presented dataset is crucial for advancing research in network traffic classification.
• CESNET-TLS-Year22 supports the development of more resilient and accurate ML-based TC solutions.
• This resource will aid in understanding and managing traffic in dynamic ISP environments.