CESNET-TimeSeries24: Time Series Dataset for Network Traffic Anomaly Detection and Forecasting

Area of Science:

• Computer Science
• Network Security
• Data Science

Background:

• Anomaly detection in network traffic is vital for cybersecurity.
• Forecasting-based methods are common, but lack extensive real-world datasets for validation.
• Existing datasets may overestimate algorithm performance, creating a false sense of progress.

Purpose of the Study:

• To address the lack of comprehensive real-world network traffic datasets.
• To provide a challenging dataset for developing and evaluating forecasting and anomaly detection models.
• To enable more realistic assessments of anomaly detection algorithm performance in ISP environments.

Main Methods:

• Collected 40 weeks of network traffic data from the CESNET3 network.
• Included traffic from 275,000 active IP addresses, reflecting diverse ISP network behavior.
• Dataset captures realistic variability for model development and evaluation.

Main Results:

• A comprehensive dataset reflecting real-world ISP network variability is now available.
• The dataset facilitates the development of more robust forecasting and anomaly detection models.
• Enables evaluations closer to practical deployment scenarios for network security.

Conclusions:

• The introduced dataset is crucial for advancing anomaly detection research.
• It allows for more accurate performance evaluations of forecasting-based security methods.
• Promotes realistic development and deployment of network security solutions.