Project AURIONAI Use-Cases Research & Innovation

TSB-AD

Time-series anomaly detection, 1,070 series from 40 datasets (VUS-PR); NeurIPS 2024; open to community submissions from 1 Apr 2026

time seriesactive

Score in VUS-PR

  1. 1Sungkyunkwan UniversityPaAno0.53 VUS-PRvendor-reported
  2. 2(Sub)-PCA0.42 VUS-PRindependent
  3. 3Carnegie Mellon UniversityMOMENT (MOMENT-1-large)0.39 VUS-PRindependent
8scores
7models
4independent
1maintainer-run
3vendor-reported

Definition

Version
not in the notes
Category
time series
Maintainer
TheDatumOrg
Status
active
Home page
https://github.com/TheDatumOrg/TSB-AD

Sources

  1. arxiv.org/pdf/2602.01359 https://arxiv.org/pdf/2602.01359
  2. GitHub - thedatumorg/TSB-AD: Time-Series Anomaly Detection | Algorithms + Datasets + Tutorials github.com https://github.com/TheDatumOrg/TSB-AD

Scores at a Glance

Score in VUS-PR

Higher is better for this unit.

1Sungkyunkwan UniversityPaAnoSungkyunkwan UniversityUnivariate (TSB-AD-U)
0.53 VUS-PR
vendor-reported2(Sub)-PCAUnivariate; as reported in the PaAno paper
0.42 VUS-PR
independent3Carnegie Mellon UniversityMOMENT (MOMENT-1-large)Carnegie Mellon UniversityUnivariate, fine-tuned (FT); as reported in the PaAno paper
0.39 VUS-PR
independent4DeepAnTMultivariate; tie with PCA; PaAno paper
0.31 VUS-PR
independent5OmniAnomalyMultivariate
0.31 VUS-PR
maintainer-run6PCAMultivariate; tie with DeepAnT; PaAno paper
0.31 VUS-PR
independent
00.20.40.6
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.

Scores

8 of 8 scores shown tsb-ad-scores.csv
ModelScoreWho ran itSettingView or populationDateSource
Sungkyunkwan UniversityPaAno0.53 VUS-PRvendor-reportedUnivariate (TSB-AD-U)rank 1, –1 Jul 2026arxiv.org
Sungkyunkwan UniversityPaAno0.43 VUS-PRvendor-reportedMultivariate (TSB-AD-M)rank 1, –1 Jul 2026arxiv.org
(Sub)-PCA0.42 VUS-PRindependentUnivariate; as reported in the PaAno paperrank 2, –1 Jul 2026arxiv.org
Carnegie Mellon UniversityMOMENT (MOMENT-1-large)0.39 VUS-PRindependentUnivariate, fine-tuned (FT); as reported in the PaAno paperrank 5, –1 Jul 2026arxiv.org
DeepAnT0.31 VUS-PRindependentMultivariate; tie with PCA; PaAno paper–1 Jul 2026arxiv.org
OmniAnomaly0.31 VUS-PRmaintainer-runMultivariate––github.com
PCA0.31 VUS-PRindependentMultivariate; tie with DeepAnT; PaAno paper–1 Jul 2026arxiv.org
IBMGranite TSPulse– VUS-PRvendor-reportedIBM claims #1 on univariate and multivariate, +20%; no scores published on card; conflicts with PaAno rank 1rank 1, ––huggingface.co
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "TSB-AD", https://aurion.semai.id/benchmarks/tsb-ad/, records of October 2026.