Overview

The CESNET-TimeSeries24 dataset [1] captures comprehensive network traffic data from the ISP network CESNET3. The dataset offers multivariate time series created through traffic aggregation at three distinct temporal intervals, providing a robust foundation for network traffic forecasting, anomaly detection, and deep learning model evaluation.

Dataset Metadata

PropertyValue
TypeOriginal dataset
CategoryTime Series
Source NetworkCESNET3 ISP
Time Span40 weeks (October 2023 – July 2024)
Aggregation Intervals10 minutes, 1 hour, 1 day
Hierarchical LevelsOverall network, 283 institutions, 548 subnets, 270,000+ IP addresses

Supported Tasks

  • Network Traffic Forecasting
  • Anomaly Detection
  • Representation Learning
  • Device Type Classification

Dataset Structure

Temporal Granularity

The dataset provides multivariate time series at three aggregation levels:

IntervalOriginal CaptureAdapted MetricsUse Cases
10 minutesYesDirect metricsFine-grained analysis, short-term forecasting
1 hourDerivedSum, average, std devMedium-term patterns, hourly trends
1 dayDerivedSum, average, std devLong-term trends, daily patterns

Note: The original capture interval was 10 minutes. For the 1-hour and 1-day series, metrics representing unique values were adapted using sum, average, and standard deviation statistics to enhance temporal granularity.

Hierarchical Coverage

The dataset’s extensive hierarchical structure enables multi-level analysis:

LevelCountResearch Applications
Overall Network1Network-wide trend analysis
Institutions283Institutional behavior patterns
Institutional Subnets548Subnet-level traffic analysis
Individual IP Addresses270,000+Device-level behavior classification

This hierarchical organization provides a robust basis for comparative analysis of deep learning models, enabling researchers to benchmark forecasting performance across multiple network levels.

Device Type Labels

The CESNET-TimeSeries24 dataset includes Device Type labels for a subset of time series, enabling device classification based on time-series behavior.

Label Hierarchy

Labels are organized in two levels:

LevelDescriptionClasses
GroupCoarse-grained categorizationserver, net-device, end-device
Group ClassFine-grained categorizationWeb server, DNS server, router, switch, etc.

Example: The group server contains more specific classes such as web-server and dns-server.

Research Applications

The 40-week temporal span and hierarchical structure make CESNET-TimeSeries24 ideal for:

  • Model Scalability Testing: Evaluate how models perform across different network scales
  • Adaptability Assessment: Test model performance across diverse traffic patterns
  • Long-term Trend Analysis: Study seasonal patterns and temporal evolution
  • Fine-grained Behavior Analysis: Examine fluctuations at multiple time scales
  • Hierarchical Forecasting: Compare performance across network hierarchy levels

How to Cite

@article{koumar2025cesnet,
  title={CESNET-TimeSeries24: Time Series Dataset for Network Traffic Anomaly Detection and Forecasting},
  author={Koumar, Josef and Hynek, Karel and {\v{C}}ejka, Tom{\'a}{\v{s}} and {\v{S}}i{\v{s}}ka, Pavel},
  journal={Scientific Data},
  volume={12},
  number={1},
  pages={338},
  year={2025},
  publisher={Nature Publishing Group UK London}
}

Download

[1] Koumar, J., Hynek, K., Čejka, T., & Šiška, P. (2024). CESNET-TimeSeries24: Time Series Dataset for Network Traffic Anomaly Detection and Forecasting [Data set]. Zenodo.
DOI: 10.5281/zenodo.13382427