Overview
The CESNET-USTS23 dataset [1] was created to evaluate the characteristics of Unevenly Sampled Time Series (USTS) from network traffic. Traffic was captured on the national CESNET2 network over a three-month period, with all IP addresses anonymized for privacy protection.
Dataset Metadata
| Property | Value |
|---|---|
| Type | Original dataset |
| Category | Time Series |
| Primary Task | Statistical Analysis of Network Traffic |
| Source Network | CESNET2 (national research and educational network) |
| Collection Period | February 2023 – April 2023 (3 months) |
| Anonymization | All IP addresses anonymized |
Unevenly Sampled Time Series
Definition
Unevenly Sampled Time Series (USTS) are temporal sequences where observations occur at irregular intervals rather than fixed timestamps. In network traffic, this naturally occurs because:
- Packets arrive at unpredictable times
- Flow durations vary significantly
- Network events are asynchronous
- Traffic patterns are bursty and irregular
Challenges for Analysis
Traditional time-series methods assume evenly spaced observations. USTS require specialized approaches:
- Interpolation Methods: Converting irregular to regular sampling
- Event-based Analysis: Working directly with irregular timestamps
- Specialized Algorithms: Time-series methods designed for irregular data
- Statistical Techniques: Adapting classical methods for uneven sampling
Time Series Types
The CESNET-USTS23 dataset contains three types of time series:
1. Packet Time Series
| Aspect | Description |
|---|---|
| Granularity | Individual packet level |
| Observations | Per-packet attributes (size, timestamp, flags) |
| Sampling | Highly irregular (packet arrival times) |
| Use Cases | Fine-grained traffic analysis, burst detection |
2. Single Flow Time Series
| Aspect | Description |
|---|---|
| Granularity | Within a single network flow |
| Observations | Packet sequences within one connection |
| Sampling | Irregular within flow duration |
| Use Cases | Flow behavior analysis, application fingerprinting |
3. Flow Time Series
| Aspect | Description |
|---|---|
| Granularity | Aggregated flow level |
| Observations | Flow-level metrics over time |
| Sampling | Irregular based on flow termination |
| Use Cases | Long-term pattern detection, dependency analysis |
Research Applications
The CESNET-USTS23 dataset enables research in:
- USTS Statistical Properties: Characterizing irregular sampling patterns in network traffic
- Method Comparison: Evaluating different approaches to handle uneven sampling
- Feature Engineering: Developing features robust to irregular timestamps
- Algorithm Development: Creating specialized methods for network USTS
- Benchmark Evaluation: Standardized testing of USTS analysis techniques
Relationship to Associated Publication
This dataset supports the research presented in:
“Unevenly Spaced Time Series from Network Traffic”
The publication explores:
- Statistical characteristics of network-derived USTS
- Challenges in applying traditional time-series methods
- Novel approaches for irregular network traffic analysis
- Performance comparisons of different USTS techniques
How to Cite
@inproceedings{koumar2023unevenly,
title={Unevenly Spaced Time Series from Network Traffic},
author={Koumar, Josef and Hynek, Karel and {\v{C}}ejka, Tom{\'a}{\v{s}}},
booktitle={2023 19th International Conference on Network and Service Management (CNSM)},
pages={1--7},
year={2023},
organization={IEEE}
}
Download
[1] Koumar, J., & Čejka, T. (2023). CESNET-USTS23: A Benchmark Dataset of Unevenly Spaced Time Series from Network Traffic (1.0) [Data set]. Zenodo.
DOI: 10.5281/zenodo.7923745