---
license: cc-by-4.0
tags:
  - network-traffic
  - time-series
  - anomaly-detection
size_categories:
  - 100M<n<1B
---

# CESNET-TimeSeries24 — verbatim Zenodo archives

A faithful mirror of [CESNET-TimeSeries24](https://zenodo.org/records/13382427)
(DOI 10.5281/zenodo.13382427): per-IP aggregate time series from 40 weeks of
CESNET backbone monitoring — 275k+ IP addresses, 548 institution subnets, 283
institutions, at 10-minute / 1-hour / 1-day windows, derived from 66 billion
flows.

**These are the upstream tar.gz archives, byte-identical, not a conversion.**
Unlike the flow corpora published alongside this mirror, TimeSeries24 holds
aggregates, not flow records — mapping it onto a flow schema would fabricate
flows that were never measured. It is published here unmodified so the whole
training corpus lives in one place; a dedicated time-series loader is the
right way to consume it.

| file | GB |
|---|---|
| ip_addresses_full.tar.gz | 40.0 |
| institution_subnets.tar.gz | 0.77 |
| institutions.tar.gz | 0.48 |
| ip_addresses_sample.tar.gz | 0.17 |
| ids_relationship.csv / times.tar.gz / weekends_and_holidays.csv | <0.01 |

Archives contain one CSV per time-series ID (metrics: `n_flows, n_packets,
n_bytes, n_dest_ip, n_dest_asn, n_dest_ports, tcp_udp_ratio_*, dir_ratio_*,
avg_duration, avg_ttl`; the 1h/1d re-aggregations carry `sum_/avg_/std_`
variants). `ts24-manifest.json` records the download provenance.

## Licensing & citation

Source dataset © CESNET, **CC-BY-4.0**; mirrored under the same license.
Cite: Koumar, J. et al., *CESNET-TimeSeries24: Time Series Dataset for
Network Traffic Anomaly Detection and Forecasting*, Scientific Data 12:338
(2025). https://doi.org/10.1038/s41597-025-04603-x