Duyu/Time-Series-Forecasting-Benchmark-Datasets

Time Series Forecasting Benchmark Datasets

Documentation Language

简体中文 | English | Tiếng Việt

Dataset Download

• https://huggingface.co/datasets/Duyu/Time-Series-Forecasting-Benchmark-Datasets/tree/main

• https://github.com/duyu09/TimeSeries-Forecasting-Dataset/releases/download/v1.0.0/dataset.7z

Dataset Desc.

• ETT The Electricity Transformer Temperature (ETT) dataset serves as a critical benchmark for evaluating electric power forecasting. It comprises two years of data collected from two separate counties in China. To analyze the impact of temporal granularity, the dataset is divided into four subsets with different sampling frequencies: ETTh1 and ETTh2 are sampled at 1-hour intervals, while ETTm1 and ETTm2 are sampled at 15-minute intervals. Each data point contains six power load-related features along with a target variable, oil temperature.
• ECL The Electricity dataset includes hourly electricity consumption data from 370 clients, providing insights into consumer-level load patterns. Data is collected from 1st January, 2011 with a sampling interval of 15 minutes.
• Weather Weather dataset consists of one year of meteorological measurements recorded every 10 minutes across 21 weather stations of the Max Planck Biogeochemistry Institute in Germany. It includes 21 variables such as air temperature, humidity, and wind speed, etc.
• Exchange Exchange comprises daily exchange rate records from 1990 to 2016 for eight foreign currencies, including those of Australia, the United Kingdom, China, Japan, Canada, Singapore, Switzerland, and New Zealand. The data is sampled at a one-day interval.
• ILI The Influenza-like Illness (ILI) dataset captures the weekly number of reported cases involving severe influenza symptoms with complications.
• Electricity This dataset represents the hourly electricity consumption of 321 clients from 2012 to 2014, measured in kilowatts (kW). It was originally extracted from the UCI repository.
• Solar This dataset contains 137 time series representing hourly solar power production in the state of Alabama throughout 2006.
• Wind This dataset contains a single extensive daily time series representing wind power production (in megawatts) recorded at 4-second intervals starting from August 1, 2019. It was downloaded from the Australian Energy Market Operator (AEMO) online platform.
• Traffic This dataset contains 15 months of daily data (440 daily records) describing the occupancy rate (between 0 and 1) of different car lanes on San Francisco Bay Area freeways over time.
• Taxi This dataset contains spatio-temporal traffic time series of New York City taxi rides recorded at 1,214 locations every 30 minutes during January 2015 and January 2016.
• Pedestrian This dataset contains hourly pedestrian counts captured from 66 sensors in Melbourne starting from May 2009. The original dataset is regularly updated when new observations become available. The dataset used here contains pedestrian counts up to April 30, 2020.
• Air-Quality This dataset was used in the KDD Cup 2018 forecasting competition. It contains hourly air quality measurements from 59 stations in two cities: Beijing (35 stations) and London (24 stations) from January 1, 2017, to March 31, 2018. The air quality measurements include various metrics such as PM2.5, PM10, NO2, CO, O3, and SO2. Missing values were imputed using leading zeros or the Last Observation Carried Forward (LOCF) method.
• Temperature This dataset contains 32,072 daily time series with temperature observations and rain forecasts gathered by the Australian Bureau of Meteorology from 422 weather stations across Australia between May 2, 2015, and April 26, 2017. Missing values were replaced with zeros, and the mean temperature column was extracted for use.
• Rain This dataset focuses on rain data extracted from the same source as the Temperature dataset.
• NN5 This dataset was used in the NN5 forecasting competition. It contains 111 time series from the banking domain with the goal of predicting daily cash withdrawals from ATMs in the UK. Missing values were replaced by the median across all corresponding days of the week throughout the entire series.
• Fred-MD This dataset contains 107 monthly time series reflecting various macroeconomic indicators sourced from the Federal Reserve Bank’s FRED-MD database. The series have been differenced and log-transformed following established practices in the literature.
• Web This dataset was used in the Kaggle Wikipedia Web Traffic forecasting competition. It contains 145,063 daily time series representing the number of hits or web traffic for Wikipedia pages from July 1, 2015, to September 10, 2017. Missing values were replaced with zeros.
• M4 The M4 dataset comprises 100,000 time series from various domains, used in the fourth Makridakis forecasting competition (M4 Competition) organized by Spyros Makridakis. It includes series with multiple frequencies—yearly, quarterly, monthly, weekly, daily, and hourly—along with an Info file providing details such as series ID, category, frequency, forecast horizon, seasonal period, and starting date.