Time Series Forecasting

MuJoCo (multi-joint dynamics with contact) is a physics engine used to implement environments to benchmark Reinforcement Learning methods.

The Medical Information Mart for Intensive Care III (MIMIC-III) dataset is a large, de-identified and publicly-available collection of medical records. Each record in the dataset includes ICD-9 codes, which identify diagnoses and procedures performed. Each code is partitioned into sub-codes, which often include specific circumstantial details. The dataset consists of 112,000 clinical reports records (average length 709.3 tokens) and 1,159 top-level ICD-9 codes. Each report is assigned to 7.6 codes, on average. Data includes vital signs, medications, laboratory measurements, observations and notes charted by care providers, fluid balance, procedure codes, diagnostic codes, imaging reports, hospital length of stay, survival data, and more. The database supports applications including academic and industrial research, quality improvement initiatives, and higher education coursework. Source: MIT Laboratory for Computational Biology

The Electricity Transformer Temperature (ETT) is a crucial indicator in the electric power long-term deployment. This dataset consists of 2 years data from two separated counties in China. To explore the granularity on the Long sequence time-series forecasting (LSTF) problem, different subsets are created, {ETTh1, ETTh2} for 1-hour-level and ETTm1 for 15-minutes-level. Each data point consists of the target value ”oil temperature” and 6 power load features. The train/val/test is 12/4/4 months. Source: https://arxiv.org/pdf/2012.07436.pdf Image Source: https://github.com/zhouhaoyi/ETDataset

Daily exchange rates of eight countries’ currencies against the US dollar, spanning from 1990 to 2010 with 7588 timesteps in total. | Dataset | Dim | Lengths | Granularity | Split | p-value | |----------------|-----|---------|-------------|--------|---------| | Exchange Rate | 8 | 7,588 | 1 day | 7:1:2 | 0.416 |

The M4 dataset is a collection of 100,000 time series used for the fourth edition of the Makridakis forecasting Competition. The M4 dataset consists of time series of yearly, quarterly, monthly and other (weekly, daily and hourly) data, which are divided into training and test sets. The minimum numbers of observations in the training test are 13 for yearly, 16 for quarterly, 42 for monthly, 80 for weekly, 93 for daily and 700 for hourly series. The participants were asked to produce the following numbers of forecasts beyond the available data that they had been given: six for yearly, eight for quarterly, 18 for monthly series, 13 for weekly series and 14 and 48 forecasts respectively for the daily and hourly ones. The M4 dataset was created by selecting a random sample of 100,000 time series from the ForeDeCk database. The selected series were then scaled to prevent negative observations and values lower than 10, thus avoiding possible problems when calculating various error measures. The scaling was performed by simply adding a constant to the series so that their minimum value was equal to 10 (29 occurrences across the whole dataset). In addition, any information that could possibly lead to the identification of the original series was removed so as to ensure the objectivity of the results. This included the starting dates of the series, which did not become available to the participants until the M4 had ended. Source: The M4 competition: Results, findings, conclusion and way forward Image Source: Are forecasting competitions data representative of the reality?

This dataset contains the traffic data in San Bernardino from July to August in 2016, with 170 detectors on 8 roads with a time interval of 5 minutes. This dataset is popular as a benchmark traffic forecasting dataset.

PeMS04 is a traffic forecasting benchmark.

METR-LA is a dataset for traffic prediction.

PEMS-BAY is a dataset for traffic prediction.

Abstract: Measurements of electric power consumption in one household with a one-minute sampling rate over a period of almost 4 years. Different electrical quantities and some sub-metering values are available. | Data Set Characteristics | Number of Instances | Area | Attribute Characteristics | Number of Attributes | Date Donated | Associated Tasks | Missing Values | | ------------------------- | ------------------- | -------- | ------------------------- | -------------------- | ------------ | ---------------------- | -------------- | | Multivariate, Time-Series | 2075259 | Physical | Real | 9 | 2012-08-30 | Regression, Clustering | Yes | ### Source: Georges Hebrail (georges.hebrail '@' edf.fr), Senior Researcher, EDF R&D, Clamart, France Alice Berard, TELECOM ParisTech Master of Engineering Internship at EDF R&D, Clamart, France ### Data Set Information: This archive contains 2075259 measurements gathered in a house located in Sceaux (7km of Paris, France) between December 2006 and November 2010 (47 months). Notes: 1. (global_active_power*1000/60 - sub_metering_1 - sub_metering_2 - sub_metering_3) represents the active energy consumed every minute (in watt hour) in the household by electrical equipment not measured in sub-meterings 1, 2 and 3. 2. The dataset contains some missing values in the measurements (nearly 1,25% of the rows). All calendar timestamps are present in the dataset but for some timestamps, the measurement values are missing: a missing value is represented by the absence of value between two consecutive semi-colon attribute separators. For instance, the dataset shows missing values on April 28, 2007. ### Attribute Information: 1. date: Date in format dd/mm/yyyy 2. time: time in format hh:mm:ss 3. global_active_power: household global minute-averaged active power (in kilowatt) 4. global_reactive_power: household global minute-averaged reactive power (in kilowatt) 5. voltage: minute-averaged voltage (in volt) 6. global_intensity: household global minute-averaged current intensity (in ampere) 7. sub_metering_1: energy sub-metering No. 1 (in watt-hour of active energy). It corresponds to the kitchen, containing mainly a dishwasher, an oven and a microwave (hot plates are not electric but gas powered). 8. sub_metering_2: energy sub-metering No. 2 (in watt-hour of active energy). It corresponds to the laundry room, containing a washing-machine, a tumble-drier, a refrigerator and a light. 9. sub_metering_3: energy sub-metering No. 3 (in watt-hour of active energy). It corresponds to an electric water-heater and an air-conditioner. ### Relevant Papers: N/A ### Citation Request: This dataset is made available under the “Creative Commons Attribution 4.0 International (CC BY 4.0)” license

Weather is recorded every 10 minutes for the 2020 whole year, which contains 21 meteorological indicators, such as air temperature, humidity, etc. The dataset in CSV format can be downloaded at https://drive.google.com/file/d/1Tc7GeVN7DLEl-RAs-JVwG9yFMf--S8dy/view?usp=share_link.

This dataset details the energy consumption of appliances in a low-energy building over 4.5 months. Data was collected at 10-minute intervals. For more information about the house, data collection, R scripts and figures, please refer to the paper and the following GitHub repository: https://github.com/LuisM78/Appliances-energy-prediction-data

PeMSD7 is traffic data in District 7 of California consisting of the traffic speed of 228 sensors while the period is from May to June in 2012 (only weekdays) with a time interval of 5 minutes. This dataset is popular for benchmark the traffic forecasting models.

PeMS07 is a traffic forecasting benchmark.

The PhysioNet Challenge 2012 dataset is publicly available and contains the de-identified records of 8000 patients in Intensive Care Units (ICU). Each record consists of roughly 48 hours of multivariate time series data with up to 37 features recorded at various times from the patients during their stay such as respiratory rate, glucose etc. Source: Multi-resolution Networks For Flexible Irregular Time Series Modeling (Multi-FIT) Image Source: https://physionet.org/content/challenge-2016/1.0.0/

The Electricity Transformer Temperature (ETT) is a crucial indicator in the electric power long-term deployment. This dataset consists of 2 years data from two separated counties in China. To explore the granularity on the Long sequence time-series forecasting (LSTF) problem, different subsets are created, {ETTh1, ETTh2} for 1-hour-level and ETTm1 for 15-minutes-level. Each data point consists of the target value ”oil temperature” and 6 power load features. The train/val/test is 12/4/4 months. Source: https://arxiv.org/pdf/2012.07436.pdf Image Source: https://github.com/zhouhaoyi/ETDataset

Abstract: The task for this dataset is to forecast the spatio-temporal traffic volume based on the historical traffic volume and other features in neighboring locations. | Data Set Characteristics | Number of Instances | Area | Attribute Characteristics | Number of Attributes | Date Donated | Associated Tasks | Missing Values | | ------------------------ | ------------------- | -------- | ------------------------- | -------------------- | ------------ | ---------------- | -------------- | | Multivariate | 2101 | Computer | Real | 47 | 2020-11-17 | Regression | N/A | ### Source: Liang Zhao, liang.zhao '@' emory.edu, Emory University. ### Data Set Information: The task for this dataset is to forecast the spatio-temporal traffic volume based on the historical traffic volume and other features in neighboring locations. Specifically, the traffic volume is measured every 15 minutes at 36 sensor locations along two major highways in Northern Virginia/Washington D.C. capital region. The 47 features include: 1) the historical sequence of traffic volume sensed during the 10 most recent sample points (10 features), 2) week day (7 features), 3) hour of day (24 features), 4) road direction (4 features), 5) number of lanes (1 feature), and 6) name of the road (1 feature). The goal is to predict the traffic volume 15 minutes into the future for all sensor locations. With a given road network, we know the spatial connectivity between sensor locations. For the detailed data information, please refer to the file README.docx. ### Attribute Information: The 47 features include: (1) the historical sequence of traffic volume sensed during the 10 most recent sample points (10 features), (2) week day (7 features), (3) hour of day (24 features), (4) road direction (4 features), (5) number of lanes (1 feature), and (6) name of the road (1 feature). ### Relevant Papers: Liang Zhao, Olga Gkountouna, and Dieter Pfoser. 2019. Spatial Auto-regressive Dependency Interpretable Learning Based on Spatial Topological Constraints. ACM Trans. Spatial Algorithms Syst. 5, 3, Article 19 (August 2019), 28 pages. DOI:[Web Link] ### Citation Request: To use these datasets, please cite the papers: Liang Zhao, Olga Gkountouna, and Dieter Pfoser. 2019. Spatial Auto-regressive Dependency Interpretable Learning Based on Spatial Topological Constraints. ACM Trans. Spatial Algorithms Syst. 5, 3, Article 19 (August 2019), 28 pages. DOI:[Web Link]

This data set contains electricity consumption of 370 points/clients.

Satellite images are snapshots of the Earth surface. We propose to forecast them. We frame Earth surface forecasting as the task of predicting satellite imagery conditioned on future weather. EarthNet2021 is a large dataset suitable for training deep neural networks on the task. It contains Sentinel~2 satellite imagery at $20$~m resolution, matching topography and mesoscale ($1.28$~km) meteorological variables packaged into $32000$ samples. Additionally we frame EarthNet2021 as a challenge allowing for model intercomparison. Resulting forecasts will greatly improve ($>\times50$) over the spatial resolution found in numerical models. This allows localized impacts from extreme weather to be predicted, thus supporting downstream applications such as crop yield prediction, forest health assessments or biodiversity monitoring. Find data, code, and how to participate at www.earthnet.tech.

VISUELLE is a repository build upon the data of a real fast fashion company, Nunalie, and is composed of 5577 new products and about 45M sales related to fashion seasons from 2016-2019. Each product in VISUELLE is equipped with multimodal information: its image, textual metadata, sales after the first release date, and three related Google Trends describing category, color and fabric popularity. Download <a href="https://drive.google.com/file/d/11Bn2efKfO_PbtdqsSqj8U6y6YgBlRcP6/view?usp=sharing">here</a> Image source: https://arxiv.org/pdf/2109.09824v1.pdf

The Electricity Transformer Temperature (ETT) is a crucial indicator in the electric power long-term deployment. This dataset consists of 2 years data from two separated counties in China. To explore the granularity on the Long sequence time-series forecasting (LSTF) problem, different subsets are created, {ETTh1, ETTh2} for 1-hour-level and ETTm1 for 15-minutes-level. Each data point consists of the target value ”oil temperature” and 6 power load features. The train/val/test is 12/4/4 months. Source: https://arxiv.org/pdf/2012.07436.pdf Image Source: https://github.com/zhouhaoyi/ETDataset

A multivariate spatio-temporal benchmark dataset for meteorological forecasting based on real-time observation data from ground weather stations.

Unified Time Series Dataset (UTSD) includes 7 domains with up to 1 billion time points with hierarchical capacities to facilitate research of large models in the field of time series. It is meticulously assembled from a blend of publicly accessible online data repositories and empirical data derived from real-world machine operations. We analyze each dataset within the collection, examining the time series through the lenses of stationarity and forecastability to allows us to characterize the level of complexity inherent to each dataset. All datasets are classified into seven distinct domains by their source: Energy, Environment, Health, Internet of Things (IoT), Nature, Transportation, and Web with diverse sampling frequencies. UTSD is constructed with hierarchical capacities, namely UTSD-1G, UTSD-2G, UTSD-4G, and UTSD-12G, where each smaller dataset is a subset of the larger ones. A larger subset means greater data difficulty and diversity, allowing you to conduct detailed scaling experiments. See the paper and codebase for more information.

The Lorenz dataset contains 100000 time-series with length 24. The data has 5 modes and it is obtained using the Lorenz equation with 5 different seed values. Source: Lorenz Dataset

Time Series Forecasting Repository containing datasets of related time series for global forecasting.

Solar Power Data for Integration Studies NREL's Solar Power Data for Integration Studies are synthetic solar photovoltaic (PV) power plant data points for the United States representing the year 2006. The data are intended for use by energy professionals—such as transmission planners, utility planners, project developers, and university researchers—who perform solar integration studies and need to estimate power production from hypothetical solar plants. Data Methodologies The Solar Power Data for Integration Studies consist of 1 year (2006) of 5-minute solar power and hourly day-ahead forecasts for approximately 6,000 simulated PV plants. Solar power plant locations were determined based on the capacity expansion plan for high-penetration renewables in Phase 2 of the Western Wind and Solar Integration Study and the Eastern Renewable Generation Integration Study. NREL generated the 5-minute data set using the Sub-Hour Irradiance Algorithm. The day-ahead solar forecast data for locations in the western United States were generated by 3TIER based on numerical weather predication simulations for Phase 1 of the Western Wind and Solar Integration Study. NREL generated the day-ahead solar forecast data in eastern U.S. locations using the Weather Research and Forecasting model. The data are for specific years and should not be assumed to be representative of typical radiation levels for a site. These data should not generally be used for site-specific project development work. Naming Convention The naming convention of the state-wise solar power data (.csv files) from the Solar Integration Studies is as follows. Data Type_Latitude_Longitude_Weather Year_PV Type_CapacityMW_Time Interval _Min.csv Data Type Actual: Real power output DA: Day ahead forecast HA4: 4 hour ahead forecast Weather Year: The PV data is based on the particular year's known weather condition. PV Type UPV: Utility scale PV DPV: Distributed PV Note: The practical difference between UPV and DPV is in the configurations (UPV has single axis tracking while DPV is fixed tilt equaling to latitude) and the smoothing (both are run through a low-pass filter, the DPV will have more of the high frequency variability smoothed out). Capacity: Installed capacity in MW Time Interval: PV generation data reading interval in minutes. Contact Yingchen Zhang Manager, Sensing, Measurement, and Forecasting Group Yingchen.Zhang@nrel.gov 303-384-7090

The data we use include 366 monthly series, 427 quarterly series and 518 yearly series. They were supplied by both tourism bodies (such as Tourism Australia, the Hong Kong Tourism Board and Tourism New Zealand) and various academics, who had used them in previous tourism forecasting studies (please refer to the acknowledgements and details of the data sources and availability). from the paper

Visuelle 2.0 is a dataset containing real data for 5355 clothing products of the retail fast-fashion Italian company, Nuna Lie. Specifically, Visuelle 2.0 provides data from 6 fashion seasons (partitioned in Autumn-Winter and Spring-Summer) from 2017-2019, right before the Covid-19 pandemic. Each product is accompanied by an HD image, textual tags and more. The time series data are disaggregated at the shop level, and include the sales, inventory stock, max-normalized prices (for the sake of confidentiality} and discounts. Exogenous time series data is also provided, in the form of Google Trends based on the textual tags and multivariate weather conditions of the stores’ locations. Finally, we also provide purchase data for 667K customers whose identity has been anonymized, to capture personal preferences. With these data, Visuelle 2.0 allows to cope with several problems which characterize the activity of a fast fashion company: new product demand forecasting, short-observation new product sales forecasting, and product recommendation.

The AQI dataset is collected from 12 observing stations around Beijing from year 2013 to 2017. The data is accessible at The University of California, Irvine (UCI) Machine Learning Repository.

Three-dimensional position of external markers placed on the chest and abdomen of healthy individuals breathing during intervals from 73s to 222s. The markers move because of the respiratory motion, and their position is sampled at approximately 10Hz. Markers are metallic objects used during external beam radiotherapy to track and predict the motion of tumors due to breathing for accurate dose delivery. The same data was used and described in detail in the following article: Krilavicius, Tomas, et al. “Predicting Respiratory Motion for Real-Time Tumour Tracking in Radiotherapy.” ArXiv:1508.00749 [Physics], Aug. 2015. arXiv.org, http://arxiv.org/abs/1508.00749.

A new spatio-temporal benchmark dataset (Hurricane), is suited for forecasting during extreme events and anomalies. The dataset is provided through the Florida Department of Revenue which provides the monthly sales revenue (2003-2020) for the tourism industry for all 67 counties of Florida which are prone to annual hurricanes. Furthermore, we aligned and joined the raw time series with the history of hurricane categories (i.e., event intensities) based on time for each county. Note that the hurricane category indicates the maximum sustained wind speed which can result in catastrophic damages as this number goes up (Category 1-6).

This repository contains a financial-domain-focused dataset for financial sentiment/emotion classification and stock market time series prediction. It's based on our paper: StockEmotions: Discover Investor Emotions for Financial Sentiment Analysis and Multivariate Time Series accepted by AAAI 2023 Bridge (AI for Financial Services). - Data collection period: Jan 2020 - Dec 2020 - Number of Utterance: 10,000 (train 80%, val 10%, test 10%) - Sentiment classes: 2 [bullish (~positive), bearish (~negative)] - Emotion classes: 12 [ambiguous, amusement, anger, anxiety, belief, confusion, depression, disgust, excitement, optimism, panic, surprise] - tweet/processed.csv: 50,281 samples with text-processed data for Topic Modelling - tweet/train, val, test.csv: 10,000 samples in total. Each file has id, date, ticker, emo_label, senti_lable, original, and processed content. For the data curation, processing (e.g. emoji, CTAG, HTAG), and annotation, we refer to our paper. The dataset is used for Financial Sentiment/Emotion Classification tasks. - price/38 companies: historical price data in csv format. The tweet and price dataset together are used for Multivariate Time Series tasks.

Graph Neural Networks (GNNs) have gained traction across different domains such as transportation, bio-informatics, language processing, and computer vision. However, there is a noticeable absence of research on applying GNNs to supply chain networks. Supply chain networks are inherently graphlike in structure, making them prime candidates for applying GNN methodologies. This opens up a world of possibilities for optimizing, predicting, and solving even the most complex supply chain problems. A major setback in this approach lies in the absence of real-world benchmark datasets to facilitate the research and resolution of supply chain problem using GNNs. To address the issue, we present a real-world benchmark dataset for temporal tasks, obtained from one of the leading FMCG companies in Bangladesh, focusing on supply chain planning for production purposes. The dataset includes temporal data as node features to enable sales predictions, production planning, and the identification of factory issues. By utilizing this dataset, researchers can employ GNNs to address numerous supply chain problems, thereby advancing the field of supply chain analytics and planning. - Dataset GitHub - arXiv - PDF on arXiv Read the paper to learn more details and data statistics.

The dataset contains historical financial transactions, including time, category and cost fields. There are 50000 clients, 205 categories and 43.7M events. The original goal was to predict the age group of the client. In this variant of the dataset, the goal is to forecast multiple future events.

The dataset includes time-stamped user product reviews behavior from January, 2008 to October, 2018. Each user has a sequence of produce review events with each event containing the timestamp and category of the reviewed product, with each category corresponding to an event type.

State-level data for the US economy through the lens of consumer spending (Credit/Debit Spending) . The dataset is enriched with state-level Economic Dynamics and Policy Responses. Specifically, we further enriched the data with the state-level policies as an indication of extreme events (e.g., the state’s business closure order). - Date: The date of the data record. - StateAbbr: Abbreviation of the state. - Population: Population of the state. - Spend_xxx: Various spending metrics (e.g., spend_acf, spend_aer, spend_all). Spend_all (all categories) - Policy: Indicator of policy measures. - lnDailyNewDeaths, DailyNewDeaths: Logarithm and raw counts of daily new deaths. - lnDailyNewCases, DailyNewCases: Logarithm and raw counts of daily new cases. - Deaths, Cases: Cumulative counts. - Day, Day_, DayOfYear: Day-related information.