A comparison of AdaBoost algorithms for time series forecast combination

Devon Barrow, S. Crone

Research output: Contribution to journalArticle

10 Citations (Scopus)
59 Downloads (Pure)

Abstract

Recently, combination algorithms from machine learning classification have been extended to time series regression, most notably seven variants of the popular AdaBoost algorithm. Despite their theoretical promise their empirical accuracy in forecasting has not yet been assessed, either against each other or against any established approaches of forecast combination, model selection, or statistical benchmark algorithms. Also, none of the algorithms have been assessed on a representative set of empirical data, using only few synthetic time series. We remedy this omission by conducting a rigorous empirical evaluation using a representative set of 111 industry time series and a valid and reliable experimental design. We develop a full-factorial design over derived Boosting meta-parameters, creating 42 novel Boosting variants, and create a further 47 novel Boosting variants using research insights from forecast combination. Experiments show that only few Boosting meta-parameters increase accuracy, while meta-parameters derived from forecast combination research outperform others.

Publisher Statement: NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006

© 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/
Original languageEnglish
Pages (from-to)1103–1119
JournalInternational Journal of Forecasting
Volume32
Issue number4
Early online date1 Jun 2016
DOIs
Publication statusPublished - Oct 2016

Fingerprint

Forecast combination
Boosting
Remedies
Model selection
Control mechanism
Empirical evaluation
Industry
Attribution
Experimental design
Editing
Machine learning
Empirical data
Peer review
Experiment
License
Quality control
Benchmark
Combination of forecasts
Factorial design

Bibliographical note

NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006

© 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

Keywords

  • Forecasting
  • Time series
  • Boosting
  • Ensemble
  • Model combination
  • Neural networks

Cite this

A comparison of AdaBoost algorithms for time series forecast combination. / Barrow, Devon; Crone, S.

In: International Journal of Forecasting, Vol. 32, No. 4, 10.2016, p. 1103–1119.

Research output: Contribution to journalArticle

Barrow, Devon ; Crone, S. / A comparison of AdaBoost algorithms for time series forecast combination. In: International Journal of Forecasting. 2016 ; Vol. 32, No. 4. pp. 1103–1119.
@article{50e518fa61364d66b346681c209989cc,
title = "A comparison of AdaBoost algorithms for time series forecast combination",
abstract = "Recently, combination algorithms from machine learning classification have been extended to time series regression, most notably seven variants of the popular AdaBoost algorithm. Despite their theoretical promise their empirical accuracy in forecasting has not yet been assessed, either against each other or against any established approaches of forecast combination, model selection, or statistical benchmark algorithms. Also, none of the algorithms have been assessed on a representative set of empirical data, using only few synthetic time series. We remedy this omission by conducting a rigorous empirical evaluation using a representative set of 111 industry time series and a valid and reliable experimental design. We develop a full-factorial design over derived Boosting meta-parameters, creating 42 novel Boosting variants, and create a further 47 novel Boosting variants using research insights from forecast combination. Experiments show that only few Boosting meta-parameters increase accuracy, while meta-parameters derived from forecast combination research outperform others.Publisher Statement: NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006{\circledC} 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/",
keywords = "Forecasting, Time series, Boosting, Ensemble, Model combination, Neural networks",
author = "Devon Barrow and S. Crone",
note = "NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006 {\circledC} 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/",
year = "2016",
month = "10",
doi = "10.1016/j.ijforecast.2016.01.006",
language = "English",
volume = "32",
pages = "1103–1119",
journal = "International Journal of Forecasting",
issn = "0169-2070",
publisher = "Elsevier",
number = "4",

}

TY - JOUR

T1 - A comparison of AdaBoost algorithms for time series forecast combination

AU - Barrow, Devon

AU - Crone, S.

N1 - NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006 © 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

PY - 2016/10

Y1 - 2016/10

N2 - Recently, combination algorithms from machine learning classification have been extended to time series regression, most notably seven variants of the popular AdaBoost algorithm. Despite their theoretical promise their empirical accuracy in forecasting has not yet been assessed, either against each other or against any established approaches of forecast combination, model selection, or statistical benchmark algorithms. Also, none of the algorithms have been assessed on a representative set of empirical data, using only few synthetic time series. We remedy this omission by conducting a rigorous empirical evaluation using a representative set of 111 industry time series and a valid and reliable experimental design. We develop a full-factorial design over derived Boosting meta-parameters, creating 42 novel Boosting variants, and create a further 47 novel Boosting variants using research insights from forecast combination. Experiments show that only few Boosting meta-parameters increase accuracy, while meta-parameters derived from forecast combination research outperform others.Publisher Statement: NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006© 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

AB - Recently, combination algorithms from machine learning classification have been extended to time series regression, most notably seven variants of the popular AdaBoost algorithm. Despite their theoretical promise their empirical accuracy in forecasting has not yet been assessed, either against each other or against any established approaches of forecast combination, model selection, or statistical benchmark algorithms. Also, none of the algorithms have been assessed on a representative set of empirical data, using only few synthetic time series. We remedy this omission by conducting a rigorous empirical evaluation using a representative set of 111 industry time series and a valid and reliable experimental design. We develop a full-factorial design over derived Boosting meta-parameters, creating 42 novel Boosting variants, and create a further 47 novel Boosting variants using research insights from forecast combination. Experiments show that only few Boosting meta-parameters increase accuracy, while meta-parameters derived from forecast combination research outperform others.Publisher Statement: NOTICE: this is the author’s version of a work that was accepted for publication in International Journal of Forecasting. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in International Journal of Forecasting, [32, 4, (2016)] DOI: 10.1016/j.ijforecast.2016.01.006© 2016, Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/

KW - Forecasting

KW - Time series

KW - Boosting

KW - Ensemble

KW - Model combination

KW - Neural networks

U2 - 10.1016/j.ijforecast.2016.01.006

DO - 10.1016/j.ijforecast.2016.01.006

M3 - Article

VL - 32

SP - 1103

EP - 1119

JO - International Journal of Forecasting

JF - International Journal of Forecasting

SN - 0169-2070

IS - 4

ER -