UniTED: A Unified Time Series Event Detection Repository powered by DAL

The Unified Event Detection Dataset (UniTED), powered by DAL, is a curated repository of time series datasets for the development and evaluation of event and anomaly detection methods. It primarily includes labeled datasets to enable consistent benchmarking and reproducible experiments.

This repository contains: - The R package that exposes dataset objects (see ?data topics and man/) - The data files shipped for package use (data/*.RData) - The ETL pipelines used to build those data files (etl/), separated into: - source/: original or labeled inputs used by ETL - intermediate/: zipped files and grouped lists produced during processing - final artifacts are saved under the repo root data/

We recommend accessing datasets via this package to ensure standardized structures and metadata.

---

Dataset Description

• Full dataset descriptions: https://github.com/cefet-rj-dal/united/wiki
• Example notebooks: Rmd/examples and rendered versions in examples/
• ETL documentation: one README per dataset under etl/<dataset>/README.md

---

Examples

Usage examples are organized by dataset: https://github.com/cefet-rj-dal/united/tree/main/examples/

library(united)
# Load mini dataset and fetch full version
data(A1Benchmark)
A1Benchmark <- loadfulldata(A1Benchmark)
#> Registered S3 method overwritten by 'quantmod':
#>   method            from
#>   as.zoo.data.frame zoo
#> Registered S3 methods overwritten by 'forecast':
#>   method  from 
#>   head.ts stats
#>   tail.ts stats

---

Installation

You can install the published version of UniTED package from GitHub https://github.com/cefet-rj-dal/united with:

timeout <- options()$timeout
options(timeout=1200)
devtools::install_github("cefet-rj-dal/united", upgrade="never")
options(timeout=timeout)

Bugs and new features request

https://github.com/cefet-rj-dal/united/issues

---

ETL Generation

The etl/ folder hosts dataset‑specific ETL pipelines that produce the final RData files included in the package. Each dataset folder (e.g., etl/numenta, etl/ucr_anomaly, etl/yahoo, etl/3W, etl/mit-bih, etl/gecco) contains:

• source/: the labeled inputs used by builder scripts
• intermediate/: zipped/raw conversions and grouped list artifacts
• builder scripts: numbered *.R files that specify the ETL order and outputs
• a README with origin links, field descriptions, and script references

Final artifacts are placed in the root data/ directory and documented in R/united.R. To reproduce data, run the numbered scripts inside etl/<dataset>/.

---

References

• Lima, J., Castro, H., Oliveira, L., Paixão, E., Baroni, L., Salles, R., Vargas, R., Ogasawara, E. UniTED: A Unified Time Series Event Detection Repository. Brazilian e-Science Workshop (BreSci). In: Brazilian e-Science Workshop (BreSci). SBC, DOI: 10.5753/bresci.2025.247972. Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys, 41(3), 1–58.
• Truong, C., Oudre, L., & Vayatis, N. (2020). Selective review of change point detection methods. Signal Processing, 167, 107299.
• Lavin, A., & Ahmad, S. (2015). Evaluating real-time anomaly detection algorithms — the Numenta Anomaly Benchmark. 2015 IEEE 14th ICMLA.