A curated list of public and research-accessible datasets for time-series anomaly detection, event detection, fault detection, and closely related benchmarking tasks.
The focus is on datasets with temporal structure and a plausible anomaly-detection use case. Some datasets are directly labeled for anomalies, while others are commonly used through benchmark preprocessing, rare-event labels, domain events, or fault classes.
- Selection Notes
- Univariate Datasets
- Multivariate Datasets
- Benchmark Collections
- Related Datasets
- Data Hubs and Catalogs
- Contributing
- Prefer official dataset pages, archival records, or maintained repositories over reuploads.
- Keep access requirements visible: open download, request form, login, or license restrictions.
- Include datasets that are useful for benchmarking, even if the original task is fault detection, event detection, operations monitoring, or rare-event classification.
- Avoid treating benchmark scores as directly comparable unless preprocessing, label policy, point adjustment, and metrics are aligned.
Synthetic and real Yahoo service time series released through Yahoo Webscope for anomaly-detection research.
- Access: request required through Yahoo Webscope.
- Notes: often used as a classic univariate anomaly-detection benchmark; check the Webscope terms before redistribution.
KPI time series from the AIOps Challenge series for detecting anomalies in large-scale IT operations metrics.
- Access: public GitHub repositories.
- Included editions: AIOps 2018, AIOps 2019, AIOps 2020.
- Related pages: 2018 announcement, 2019 announcement, 2020 announcement.
Competition-style time-series anomaly-detection data associated with the KDDCup2021 practice material.
- Access: HexagonML login required for the competition page.
- Public mirror: intellygenta/KDDCup2021.
Synthetic Mackey-Glass time series with deliberately injected, non-trivial anomalies.
- Access: open Zenodo record.
- Repository: MarkusThill/MGAB.
- Publication: Time Series Encodings with Temporal Convolutional Networks.
Real service and client telemetry time series from Microsoft cloud monitoring scenarios, with expert-labeled anomaly points.
- Access: public GitHub repository.
- Domain: production cloud telemetry, service rates, latencies, crash rates, and related operational metrics.
A modified subset of KDD Cup 1999 network traffic records used for outlier and attack detection.
- Access: open ODDS download page.
- Notes: tabular/network dataset with temporal-adjacent usage in anomaly-detection benchmarks; validate suitability for sequence models before use.
Chronologically ordered network-packet feature streams from IoT and surveillance-system attacks, with benign/malicious labels.
- Access: public UCI Machine Learning Repository dataset.
- Domain: online intrusion detection, IoT network monitoring, and sequential attack detection.
- Publication: Kitsune: An Ensemble of Autoencoders for Online Network Intrusion Detection.
Sequential network-traffic feature data from nine commercial IoT devices infected by Mirai and BASHLITE botnets.
- Access: public UCI Machine Learning Repository dataset.
- Domain: IoT intrusion detection and anomaly-based botnet detection.
- Publication: N-BaIoT: Network-Based Detection of IoT Botnet Attacks Using Deep Autoencoders.
Large-scale ISP traffic time series derived from the CESNET3 network for network-traffic forecasting and anomaly detection.
- Access: open Zenodo record; helper tooling is available through CESNET/cesnet-tszoo.
- Scope: 40 weeks of traffic aggregates over IP, institutional, and subnet levels.
- Publication: CESNET-TimeSeries24: Time Series Dataset for Network Traffic Anomaly Detection and Forecasting.
Curated repository linking multiple open wind-turbine SCADA datasets from different wind farms.
- Access: mixed; some sources require platform registration or an application.
- Domain: renewable-energy operations, turbine telemetry, condition monitoring, and fault detection.
Real-world SCADA data from three wind farms, released for wind-turbine anomaly and early-fault detection.
- Access: open Zenodo record.
- Publication: CARE to Compare: A real-world dataset for anomaly detection in wind turbine data.
A standard industrial process-control benchmark built around simulated plant operations and process faults.
- Access: open Harvard Dataverse record.
- Helpful resources: TEP introduction and PyTEP.
- Notes: PyTEP requires an activated MATLAB/Simulink license for customized simulations.
Server Machine Dataset (SMD)
Server-machine telemetry used for multivariate anomaly detection in operations monitoring.
- Access: public through the OmniAnomaly repository.
- Publication: Robust Anomaly Detection for Multivariate Time Series through Stochastic Recurrent Neural Network.
Water-quality sensor data from the GECCO Industrial Challenges on online anomaly and event detection for drinking-water monitoring.
- Access: open Zenodo records for GECCO 2018 and GECCO 2019; original challenge page: GECCO 2018.
- Domain: environmental IoT and water-quality event detection.
Application-server metrics from a large Internet company, released with the InterFusion benchmark material.
- Access: public GitHub repository.
- Publication: Multivariate Time Series Anomaly Detection and Interpretation using Hierarchical Inter-Metric and Temporal Embedding.
Spacecraft telemetry from NASA's Soil Moisture Active Passive satellite and Mars Science Laboratory rover.
- Access: processed benchmark data in Telemanom and OmniAnomaly; source programs: SMAP and MSL.
- Publications: Detecting Spacecraft Anomalies Using LSTMs and Nonparametric Dynamic Thresholding and OmniAnomaly.
Cyber-physical security datasets from Singapore University of Technology and Design and the iTrust Centre for Research in Cyber Security.
Secure Water Treatment (SWaT)
Water-treatment testbed data with normal operation and deliberate cyber-physical attack scenarios.
- Scope: 11 days of operation, including 7 days normal and 4 days under attack.
- Signals: 51 sensors and actuators with normal/abnormal labels.
Water Distribution (WADI)
Water-distribution testbed data with normal operation and deliberate attack scenarios.
- Scope: 16 days of operation, including 14 days normal and 2 days under attack.
- Signals: 123 sensors and actuators with attack periods.
A University of Alabama in Huntsville collection covering cyber attacks and faults in industrial-control settings.
- Access: public dataset page.
- Domains: power systems, gas pipelines, water storage tanks, and energy-management systems.
Challenge datasets with SCADA measurements from the C-Town water distribution network under normal operation and cyber-attack scenarios.
- Access: public challenge website.
- Domain: water-distribution cyber-physical attack detection.
- Publication: Battle of the Attack Detection Algorithms: Disclosing Cyber Attacks on Water Distribution Networks.
Real-time water-distribution testbed data with cyberattack, leakage, and sensor-failure events for event diagnosis and anomaly detection.
- Access: open Zenodo record.
- Domain: smart water systems, operational failures, and cyber-physical event diagnosis.
- Related work: Hardware-in-the-loop investigations of multimodal cybersecurity and operational failure detection in smart water systems.
Hourly multi-source water-distribution data from a Slovak utility, with SCADA, energy, and environmental anomaly scores mapped to confirmed leak labels.
- Access: open Zenodo record.
- Domain: urban water-network leak detection, fault prediction, and early-warning systems.
- Publication: A multisource dataset for anomaly detection and fault prediction in urban water distribution networks.
Labeled leak and no-leak sensor signals from a laboratory-scale water distribution testbed.
- Access: open Mendeley Data record.
- Signals: accelerometer, hydrophone, and dynamic-pressure measurements under multiple leak types, network topologies, and background conditions.
- Publication: Benchmarking dataset for leak detection and localization in water distribution systems.
Kaggle dataset with roughly 509k rows, 11 features, and a very low anomaly density.
- Access: Kaggle account may be required.
- Notes: useful as a realistic imbalance case; provenance is less clear than archival or paper-backed datasets.
Helicopter vibration measurements from accelerometers placed at multiple positions and directions.
- Access: ETH Research Collection record.
- Domain: aerospace vibration validation and abnormal sensor behavior detection.
Waveform time series from High Voltage Converter Modulators at the Spallation Neutron Source, labeled as normal or faulty.
- Access: open Mendeley Data record.
- Domain: particle accelerator power electronics, early fault detection, and signal-based anomaly detection.
- Publication: Real electronic signal data from particle accelerator power systems for machine learning anomaly detection.
Pooled Server Metrics (PSM)
Multivariate eBay KPI data with per-minute cart-volume metrics across business and user subdimensions.
- Access: public GitHub repository.
- Publications: Practical Approach to Asynchronous Multivariate Time Series Anomaly Detection and Localization and Real-Time Synchronization in Neural Networks for Multivariate Time Series Anomaly Detection.
Oil-well time-series instances containing rare undesirable events, released with tooling for dataset exploration and experimentation.
- Access: public GitHub repository.
- Publication: A realistic and public dataset with rare undesirable real events in oil wells.
Chemical-process time-series anomaly-detection collection with industrial, pilot-scale, and laboratory-scale process data.
- Access: public Kaggle dataset; code repository: wagner-d/noboom.
- Domain: chemical process monitoring and industrial fault detection.
- Publication: NoBOOM: Chemical Process Datasets for Industrial Anomaly Detection.
Industrial packaging-machine production intervals, machine states, alarms, throughput, and one-hour aggregate sequences.
- Access: open Zenodo record.
- Domain: packaging machinery, alarm forecasting, throughput monitoring, and industrial anomaly detection.
Air-compressor sensor data from the Porto metro system for online anomaly detection and failure prediction in railway maintenance.
- Access: public UCI Machine Learning Repository dataset; archival record: Zenodo.
- Domain: predictive maintenance, air-production units, and metro operations.
- Publication: The MetroPT dataset for predictive maintenance.
Bearing vibration data for normal bearings and single-point drive-end and fan-end defects.
- Access: public download page.
- Domain: rotating machinery and fault diagnosis.
Gearbox Faults (LASPI)
Current, voltage, and vibration measurements from an electromechanical drive system with gearbox faults.
- Access: UBFC data record.
- Domain: gearbox fault detection and diagnosis.
Speed, current, voltage, and vibration measurements from a three-phase asynchronous motor system.
- Access: UBFC data record.
- Domain: rotating-machine fault detection and diagnosis.
Industrial-control testbed data augmented with Hardware-in-the-Loop simulation for power generation and pumped-storage hydropower scenarios.
- Access: public GitHub repository.
- Domain: ICS anomaly detection and cyber-physical security.
Sensor recordings from five people performing activity scenarios, commonly adapted for fall or rare-event detection.
- Access: public UCI dataset page.
- Refactored version: Kaggle anomaly-detection falling events.
Equidistant telemetry from 97 sensors in the EDEN ISS research greenhouse.
- Access: open Zenodo record.
- Publication: Unraveling Anomalies in Time.
Benchmark for explainable anomaly detection over high-dimensional time series from repeated Apache Spark executions.
- Access: public GitHub repository; dataset files are included under the repository data path.
- Publication: Exathlon: A Benchmark for Explainable Anomaly Detection over Time Series.
Multi-source observability data from an OpenStack system, including metrics, logs, traces, workloads, and injected faults.
- Access: open Zenodo record.
- Repository: SashoNedelkoski/multi-source-observability-dataset.
- Domain: AIOps anomaly detection, root-cause analysis, and multimodal operations analytics.
A large archive of univariate time-series anomaly-detection datasets designed to address common benchmark flaws.
- Access: direct ZIP download; alternative record: Figshare.
- Publication: Current Time Series Anomaly Detection Benchmarks are Flawed and are Creating the Illusion of Progress.
Open benchmark suite for evaluating univariate time-series anomaly-detection methods across many labeled series and domains.
- Access: public GitHub repository.
- Related repository: TheDatumOrg/TSB-AD.
Skoltech Anomaly Benchmark (SKAB)
Multivariate benchmark built from sensor readings of a testbed under different anomaly and changepoint scenarios.
- Access: public GitHub repository.
- Includes: datasets, evaluation utilities, examples, and benchmark tooling.
Streaming anomaly-detection benchmark with labeled real and synthetic time series plus a scoring system for real-time detection.
- Access: public GitHub repository; overview page: Numenta NAB resource.
- Notes: useful for online detection experiments; scoring assumptions differ from pointwise offline benchmarks.
Simulated multivariate benchmark with controlled anomaly injection and detailed ground truth.
- Access: open Zenodo record.
- Scope: 17 variables, 5 million timestamps, and 200 injected anomalies.
Synthetic time-series anomaly generator and dataset collection integrated with TimeEval.
- Access: public GitHub repository and Python package.
- Scope: configurable univariate and multivariate time series with multiple anomaly kinds.
- Publication: TimeEval: A Benchmarking Toolkit for Time Series Anomaly Detection Algorithms.
Dataset overview behind the TimeEval evaluation work, covering many univariate and multivariate anomaly-detection datasets.
- Access: public overview and linked downloads.
- Tooling: TimeEval.
- Notes: a useful cross-check for dataset provenance, dimensions, labels, and benchmark metadata.
Large-scale benchmark for multivariate time-series anomaly detection and model selection.
- Access: public GitHub repository with dataset-loading guidance.
- Publication: mTSBench: Benchmarking Multivariate Time Series Anomaly Detection and Model Selection at Scale.
These datasets are not always direct anomaly-detection benchmarks, but they are useful for rare-event detection, temporal outlier detection, event detection, or realistic preprocessing examples.
Passenger-count time series with known unusual events such as holidays, the NYC marathon, and a snow storm.
Hourly traffic volume on Interstate 94 with weather and holiday covariates.
Chronologically ordered credit-card transactions with highly imbalanced fraud labels.
- Notes: not a classical time-series dataset, but useful for temporal fraud-detection baselines.
Synthetic mobile-money transaction data for fraud-detection experiments.
- Related Kaggle dataset: PaySim 1.0.
Wearable accelerometer recordings from Parkinson's disease patients with annotated freezing-of-gait episodes.
- Access: public UCI dataset page.
- Domain: wearable health monitoring and rare-event detection in multivariate sensor streams.
Two-stream people-count data from the UCI CalIt2 building, used for detecting unusual building events.
- Access: public UCI dataset page.
- Domain: urban/building event detection.
Open datasets for solar photovoltaic, wind, and thermal-energy technology.
Catalog of public time-series dataset portals, benchmark collections, and domain-specific sources.
Large collection of physiological and clinical research data with software and documentation.
Dataset sharing platform connected to IEEE research publications.
Open-science search for time-series anomaly-detection datasets across CSV, TXT, ZIP, HDF5, and XLSX records.
Community dataset hub with an increasing number of time-series and anomaly-detection dataset mirrors.
NASA repository for prognostics and systems-health datasets, including battery, bearing, milling, and C-MAPSS turbofan degradation time series.
Prognostics and Health Management Society repository for data challenges and condition-monitoring datasets.
Research dataset repository for anomaly detection across modalities, including a dedicated time-series section.
Contributions are welcome. Please prefer official sources and include enough context for readers to decide whether a dataset fits their benchmark.
See CONTRIBUTING.md for the recommended entry format and review checklist.