Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Monash Time Series Forecasting Archive

Rakshitha Godahewa, Christoph Bergmeir, Geoffrey I. Webb, Rob J. Hyndman, Pablo Montero-Manso

TL;DR

The paper addresses the lack of benchmarks for global time series forecasting by introducing the Monash Time Series Forecasting Archive, a comprehensive collection of 20 datasets (with 50 variations) spanning varied domains and frequencies, plus a flexible tsf data format and accompanying tooling. It characterizes the datasets through feature analysis using tsfeatures and catch22, and provides baseline forecast evaluations across eight error metrics for seven forecasting methods including a globally trained model. Key contributions include the dataset archive, a standardized data format, feature-space insights, and publicly available baseline results and code to enable rigorous benchmarking of future global forecasting approaches. The work is significant for practitioners and researchers aiming to develop and evaluate cross-series forecasting methods with realistic, heterogeneous data and reproducible evaluation pipelines.

Abstract

Many businesses and industries nowadays rely on large quantities of time series data making time series forecasting an important research area. Global forecasting models that are trained across sets of time series have shown a huge potential in providing accurate forecasts compared with the traditional univariate forecasting models that work on isolated series. However, there are currently no comprehensive time series archives for forecasting that contain datasets of time series from similar sources available for the research community to evaluate the performance of new global forecasting algorithms over a wide variety of datasets. In this paper, we present such a comprehensive time series forecasting archive containing 20 publicly available time series datasets from varied domains, with different characteristics in terms of frequency, series lengths, and inclusion of missing values. We also characterise the datasets, and identify similarities and differences among them, by conducting a feature analysis. Furthermore, we present the performance of a set of standard baseline forecasting methods over all datasets across eight error metrics, for the benefit of researchers using the archive to benchmark their forecasting algorithms.

Monash Time Series Forecasting Archive

TL;DR

The paper addresses the lack of benchmarks for global time series forecasting by introducing the Monash Time Series Forecasting Archive, a comprehensive collection of 20 datasets (with 50 variations) spanning varied domains and frequencies, plus a flexible tsf data format and accompanying tooling. It characterizes the datasets through feature analysis using tsfeatures and catch22, and provides baseline forecast evaluations across eight error metrics for seven forecasting methods including a globally trained model. Key contributions include the dataset archive, a standardized data format, feature-space insights, and publicly available baseline results and code to enable rigorous benchmarking of future global forecasting approaches. The work is significant for practitioners and researchers aiming to develop and evaluate cross-series forecasting methods with realistic, heterogeneous data and reproducible evaluation pipelines.

Abstract

Many businesses and industries nowadays rely on large quantities of time series data making time series forecasting an important research area. Global forecasting models that are trained across sets of time series have shown a huge potential in providing accurate forecasts compared with the traditional univariate forecasting models that work on isolated series. However, there are currently no comprehensive time series archives for forecasting that contain datasets of time series from similar sources available for the research community to evaluate the performance of new global forecasting algorithms over a wide variety of datasets. In this paper, we present such a comprehensive time series forecasting archive containing 20 publicly available time series datasets from varied domains, with different characteristics in terms of frequency, series lengths, and inclusion of missing values. We also characterise the datasets, and identify similarities and differences among them, by conducting a feature analysis. Furthermore, we present the performance of a set of standard baseline forecasting methods over all datasets across eight error metrics, for the benefit of researchers using the archive to benchmark their forecasting algorithms.

Paper Structure

This paper contains 34 sections, 6 equations, 3 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Datasets
  3. Data Format
  4. Time Series Datasets
  5. M1 Dataset
  6. M3 Dataset
  7. M4 Dataset
  8. Tourism Dataset
  9. NN5 Dataset
  10. CIF 2016 Dataset
  11. Kaggle Web Traffic Dataset
  12. Solar Dataset
  13. Electricity Dataset
  14. London Smart Meters Dataset
  15. Wind Farms Dataset
  16. ...and 19 more sections

Figures (3)

  • Figure 1: An example of the file format for the NN5 daily dataset
  • Figure 2: The directions of the 5 feature components: ACF1, trend, entropy, seasonal strength, and lambda, for the two-dimensional feature space generated by the first two principal components (PC1, PC2) extracted with PCA.
  • Figure 3: Hexbin plots showing the normalised density values of the low-dimensional feature space generated by PCA across ACF1, trend, entropy, seasonal strength and lambda for 20 datasets. The dark and light hexbins denote the high and low density areas, respectively.