Contamination MapTime series / Model
DADA (General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders)
- Organisation
- East China Normal University / Aalborg University / Huawei Noah's Ark Lab
- Family
- DADA
- Released
- 2024-05-24 (arXiv v1 submission date, https://arxiv.org/abs/2405.15273v1)
- Weights
- Unknown
- Links
- PaperCode
Training data disclosure: Full list
Level determined for 24% of benchmark cells.
- TestTest data seen2
- SeenSeen, split unclear3
- TrainTraining data seen0
- Related data seen6
- NoNot seen1
- ?Unknown37
“We collect two datasets for pre-training, namely Anomaly Detection Data and Monash+ , each of them contains normal time series and abnormal time series.”
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
“The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
“The pre-training datasets consist of approximately 400 million time points”
Other declared training data
-
ASD · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Exathlon · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
ECG · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
MITDB · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
OPP · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
SVDB · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
GAIA · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
IOPS · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
MGAB · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
NYC · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
SKAB · split: full
Anomaly Detection Data subset: training data used as normal data and labelled test data as abnormal data, so test portions were used (whether all test data or only anomalous segments is not fully clear).
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Yahoo S5 Labeled Anomaly Detection Dataset (Yahoo Webscope S5) · split: full
Yahoo S5 (Laptev et al., 2015) in DADA's Anomaly Detection Data; training data as normal and labelled test data as abnormal, i.e. labelled anomalies of Yahoo were seen in pre-training. Which Yahoo subsets (A1-A4) is not stated.
Evidence
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 “DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
arxiv.org third-party study· study · Appendix D.3 · retrieved 2026-10-15 “We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
arxiv.org third-party study· study · Appendix D.3 · retrieved 2026-10-15 -
Aus. Electricity Demand · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Wind · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Wind Farms · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Solar Power · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Solar (not yet matched to a dataset) · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
London Smart Meters · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Saugeen River Flow · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Sunspot · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Weather · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
KDD Cup 2018 · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
US Births · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
FRED MD · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
Bitcoin · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15 -
NN5 · split: unknown
Monash+ (Monash series plus synthetic anomaly-injected copies). 'Weather' here is the Monash weather dataset, not the long-horizon Jena Weather.
Evidence
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
arxiv.org · paper · Appendix, Table 6 · retrieved 2026-10-15 “The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15
Declared exclusions
-
Dataset Server Machine Dataset (SMD) · entirely excluded
Evidence
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
arxiv.org · paper · Section 4.1 / Appendix A.1 · retrieved 2026-10-15 “we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 -
Dataset Mars Science Laboratory (MSL) telemetry anomaly dataset · entirely excluded
Evidence
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
arxiv.org · paper · Section 4.1 / Appendix A.1 · retrieved 2026-10-15 “we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 -
Dataset Soil Moisture Active Passive (SMAP) telemetry anomaly dataset · entirely excluded
Evidence
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
arxiv.org · paper · Section 4.1 / Appendix A.1 · retrieved 2026-10-15 “we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 -
Dataset Secure Water Treatment (SWaT) · entirely excluded
Evidence
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
arxiv.org · paper · Section 4.1 / Appendix A.1 · retrieved 2026-10-15 “we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 -
Dataset Pooled Server Metrics (PSM) · entirely excluded
Evidence
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
arxiv.org · paper · Section 4.1 / Appendix A.1 · retrieved 2026-10-15 “we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 - Dataset UCR Time Series Anomaly Archive (Hexagon ML/UCR, KDD Cup 2021) · entirely excluded
Contamination analyses
Analyses that found benchmark test data in this model's training data, or in a corpus it uses.
-
Yahoo S5 Labeled Anomaly Detection Dataset (Yahoo Webscope S5)
· pre-training data
Third-party (TimeRCD authors): DADA's pre-training data includes IOPS, MGAB and YAHOO, which overlap their TSB-AD-based zero-shot test sets; DADA results excluded where affected.
Evidence
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
arxiv.org third-party study· study · Appendix D.3 · retrieved 2026-10-15 “We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
arxiv.org third-party study· study · Appendix D.3 · retrieved 2026-10-15 -
TSB-AD-U (TSB-AD univariate)
· pre-training data
Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.
Evidence
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
arxiv.org third-party study· study · Appendix D.3 · retrieved 2026-10-15 “We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
arxiv.org third-party study· study · Appendix D.3 · retrieved 2026-10-15 “Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
arxiv.org third-party study· study · Appendix D.1 · retrieved 2026-10-15
Benchmark cells
GIFT-Eval
Unknown Jena Weather (GIFT-Eval) · training data not disclosed in enough detail
Unknown BizITObs - Application · training data not disclosed in enough detail
Unknown BizITObs - Service · training data not disclosed in enough detail
Unknown BizITObs - L2C · training data not disclosed in enough detail
Unknown Bitbrains - Fast Storage · training data not disclosed in enough detail
Unknown Bitbrains - rnd · training data not disclosed in enough detail
Unknown Restaurant (Recruit Restaurant Visitor Forecasting) · training data not disclosed in enough detail
Unknown ETT1 (GIFT-Eval) · training data not disclosed in enough detail
Unknown ETT2 (GIFT-Eval) · training data not disclosed in enough detail
Unknown Loop Seattle (LibCity) · training data not disclosed in enough detail
Unknown SZ-Taxi (LibCity) · training data not disclosed in enough detail
Unknown M_DENSE (LibCity) · training data not disclosed in enough detail
Unknown Solar (LSTNet, 137 Alabama PV plants) · training data not disclosed in enough detail
Unknown Hierarchical Sales (Mancuso et al., 2020) · training data not disclosed in enough detail
Unknown M4 Yearly · training data not disclosed in enough detail
Unknown M4 Quarterly · training data not disclosed in enough detail
Unknown M4 Monthly · training data not disclosed in enough detail
Unknown M4 Weekly · training data not disclosed in enough detail
Unknown M4 Daily · training data not disclosed in enough detail
Unknown M4 Hourly · training data not disclosed in enough detail
Unknown Hospital · training data not disclosed in enough detail
Unknown COVID Deaths · training data not disclosed in enough detail
Seen, split unclear US Births · the dataset was used; the source does not say which part
Evidence (2)
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
“The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
Seen, split unclear Saugeen River Flow · the dataset was used; the source does not say which part
Evidence (2)
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
“The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
Unknown Temperature Rain · training data not disclosed in enough detail
Seen, split unclear KDD Cup 2018 · the dataset was used; the source does not say which part
Evidence (2)
“Aus. Electricity Demand Wind Wind Farms Solar Power Solar London Smart Meters Saugeen River Flow Sunspot Weather KDD Cup 2018 US Births FRED MD Bitcoin NN5”
“The normal time series in Monash+ comes from the Monash Prediction Library (Godahewa et al., 2021) as shown in Table 6, and the abnormal time series in Monash+ are generated from Monash by anomaly injection.”
Unknown Car Parts · training data not disclosed in enough detail
Unknown ElectricityLoadDiagrams20112014 (UCI) · training data not disclosed in enough detail
Long-horizon forecasting (ETT, Electricity, Traffic, Weather, Exchange, ILI)
Unknown ETTh1 · training data not disclosed in enough detail
Unknown ETTh2 · training data not disclosed in enough detail
Unknown ETTm1 · training data not disclosed in enough detail
Unknown ETTm2 · training data not disclosed in enough detail
Unknown Electricity (ECL, 321 clients) · training data not disclosed in enough detail
Unknown Traffic (PeMS San Francisco Bay, 862 sensors) · training data not disclosed in enough detail
Unknown Weather (Jena, 2020; Autoformer) · training data not disclosed in enough detail
Unknown Exchange-Rate · training data not disclosed in enough detail
Unknown ILI (CDC influenza-like illness, national) · training data not disclosed in enough detail
Time-series anomaly detection
Test data seen TSB-AD-U (TSB-AD univariate) · a contamination analysis found test examples in the pre-training data (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]
Evidence (3)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
Unknown TSB-AD-M (TSB-AD multivariate) · training data not disclosed in enough detail
UCR Time Series Anomaly Archive (Hexagon ML/UCR, KDD Cup 2021) · a contamination analysis found test examples of the related dataset TSB-AD-U (TSB-AD univariate) (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]; the authors state this dataset was excluded
Evidence (8)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
“UCR [106] is a collection of univariate time series of multiple domains including air temperature, arterial blood pressure, ABP, astronomy, EPG, ECG, gait, power demand, respiration, walking accelerator.”
“R. Wu and E. J. Keogh. Current time series anomaly detection benchmarks are flawed and are creating”
“303_UCR_id_1_Medical_tr_2499_1st_8000.csv”
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
“We also evaluate UCR (Wu & Keogh, 2023) in the Appendix C.2.”
Server Machine Dataset (SMD) · a contamination analysis found test examples of the related dataset TSB-AD-U (TSB-AD univariate) (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]; the authors state this dataset was excluded
Evidence (8)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
“we implement a transformation strategy that converts multivariate time series into univariate formats by treating each channel as an independent series.”
“SMD (U) [97]”
“178_SMD_id_1_Facility_tr_6000_1st_10609.csv”
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
“we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
Mars Science Laboratory (MSL) telemetry anomaly dataset · a contamination analysis found test examples of the related dataset TSB-AD-U (TSB-AD univariate) (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]; the authors state this dataset was excluded
Evidence (8)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
“we implement a transformation strategy that converts multivariate time series into univariate formats by treating each channel as an independent series.”
“MSL (U) [43]”
“140_MSL_id_1_Sensor_tr_530_1st_630.csv”
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
“we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
Soil Moisture Active Passive (SMAP) telemetry anomaly dataset · a contamination analysis found test examples of the related dataset TSB-AD-U (TSB-AD univariate) (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]; the authors state this dataset was excluded
Evidence (8)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
“we implement a transformation strategy that converts multivariate time series into univariate formats by treating each channel as an independent series.”
“SMAP (U) [43]”
“531_SMAP_id_1_Sensor_tr_1811_1st_4510.csv”
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
“we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
Not seen Pooled Server Metrics (PSM) · the authors state the dataset was excluded from training
Evidence (2)
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
“we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
Secure Water Treatment (SWaT) · a contamination analysis found test examples of the related dataset TSB-AD-U (TSB-AD univariate) (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]; the authors state this dataset was excluded
Evidence (8)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
“we implement a transformation strategy that converts multivariate time series into univariate formats by treating each channel as an independent series.”
“SWaT (U) [62]”
“550_SWaT_id_1_Sensor_tr_43700_1st_43800.csv”
“None of the downstream datasets for evaluation are included in the pre-training datasets.”
“we evaluate five widely used benchmark datasets: SMD (Su et al., 2019), MSL (Hundman et al., 2018), SMAP (Hundman et al., 2018), SWaT (Mathur & Tippenhauer, 2016), PSM (Abdulaal et al., 2021), and NeurIPS-TS (including CICIDS, Creditcard, GECCO and SWAN) (Lai et al., 2021).”
Test data seen Yahoo S5 Labeled Anomaly Detection Dataset (Yahoo Webscope S5) · the whole dataset was used
Evidence (4)
“Table 5 lists subsets in Anomaly Detection Data, including ASD (Li et al., 2021a), Exathlon (Jacob et al., 2021), ECG (Moody & Mark, 2001), MITDB (Moody & Mark, 2001), OPP (Roggen et al., 2010), SVDB (Greenwald et al., 1990), GAIA1 , IOPS (Ren et al., 2019), MGAB (Thill et al., 2020), NYC (Cui et al., 2016), SKAB2 , YAHOO (Laptev et al., 2015).”
“We take the training data of this part as normal data, and the data with anomaly labels in test data as abnormal data.”
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
Numenta Anomaly Benchmark (NAB) · a contamination analysis found test examples of the related dataset TSB-AD-U (TSB-AD univariate) (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]
Evidence (5)
“DADA: Its pre-training data includes IOPS, MGAB, and YAHOO (Shentu et al., 2024).”
“We exclude certain baseline results (specifically from DADA and MOMENT) because their pre-training corpora heavily overlap with our zero-shot test benchmarks, violating the zero-shot evaluation setting.”
“Our evaluation uses a selection of datasets from one primary source, the TSB-AD benchmark by (Liu & Paparrizos, 2024).”
“NAB [6] is composed of labeled real-world and artificial time series including AWS server metrics, online advertisement clicking rates, real-time traffic data, and a collection of Twitter mentions of large publicly-traded companies.”
“001_NAB_id_1_Facility_tr_1007_1st_2014.csv”
Time-series classification
Unknown UCR Time Series Classification Archive (2018) · training data not disclosed in enough detail
Unknown UEA Multivariate Time Series Classification Archive (2018) · training data not disclosed in enough detail
Notes
Zero-shot anomaly detector pre-trained on ~400M points: 12 anomaly-detection datasets (incl. labelled test data) + Monash with injected anomalies. Explicit exclusion of all evaluation datasets (SMD, MSL, SMAP, SWaT, PSM, NeurIPS-TS, UCR). Pre-training includes YAHOO, IOPS, MGAB, ECG, MITDB, SVDB, OPP, Exathlon - sources that are also in TSB-AD-U/-M, so TSB-AD evaluations of DADA are contaminated (as TimeRCD notes). The GitHub repo linked in the paper returns 404 (2026-10-15); no released checkpoint found.
Sources
- DADA (ICLR 2025, arXiv v4) · paper · retrieved 2026-10-15