Visualising forecasting algorithm performance using time series instance spaces

Yanfei Kang, Rob Hyndman, Kate Smith-Miles

Research output: Contribution to journalArticleResearchpeer-review

21 Citations (Scopus)

Abstract

It is common practice to evaluate the strength of forecasting methods using collections of well-studied time series datasets, such as the M3 data. The question is, though, how diverse and challenging are these time series, and do they enable us to study the unique strengths and weaknesses of different forecasting methods? This paper proposes a visualisation method for collections of time series that enables a time series to be represented as a point in a two-dimensional instance space. The effectiveness of different forecasting methods across this space is easy to visualise, and the diversity of the time series in an existing collection can be assessed. Noting that the diversity of the M3 dataset has been questioned, this paper also proposes a method for generating new time series with controllable characteristics in order to fill in and spread out the instance space, making our generalisations of forecasting method performances as robust as possible.

Original languageEnglish
Pages (from-to)345-358
Number of pages14
JournalInternational Journal of Forecasting
Volume33
Issue number2
DOIs
Publication statusPublished - 1 Apr 2017

Keywords

  • Forecasting algorithm comparison
  • M3-Competition
  • Time series generation
  • Time series visualisation

Cite this

@article{b56eb576578f47809ab63c0d4075e842,
title = "Visualising forecasting algorithm performance using time series instance spaces",
abstract = "It is common practice to evaluate the strength of forecasting methods using collections of well-studied time series datasets, such as the M3 data. The question is, though, how diverse and challenging are these time series, and do they enable us to study the unique strengths and weaknesses of different forecasting methods? This paper proposes a visualisation method for collections of time series that enables a time series to be represented as a point in a two-dimensional instance space. The effectiveness of different forecasting methods across this space is easy to visualise, and the diversity of the time series in an existing collection can be assessed. Noting that the diversity of the M3 dataset has been questioned, this paper also proposes a method for generating new time series with controllable characteristics in order to fill in and spread out the instance space, making our generalisations of forecasting method performances as robust as possible.",
keywords = "Forecasting algorithm comparison, M3-Competition, Time series generation, Time series visualisation",
author = "Yanfei Kang and Rob Hyndman and Kate Smith-Miles",
year = "2017",
month = "4",
day = "1",
doi = "10.1016/j.ijforecast.2016.09.004",
language = "English",
volume = "33",
pages = "345--358",
journal = "International Journal of Forecasting",
issn = "0169-2070",
publisher = "Elsevier",
number = "2",

}

Visualising forecasting algorithm performance using time series instance spaces. / Kang, Yanfei; Hyndman, Rob; Smith-Miles, Kate.

In: International Journal of Forecasting, Vol. 33, No. 2, 01.04.2017, p. 345-358.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Visualising forecasting algorithm performance using time series instance spaces

AU - Kang, Yanfei

AU - Hyndman, Rob

AU - Smith-Miles, Kate

PY - 2017/4/1

Y1 - 2017/4/1

N2 - It is common practice to evaluate the strength of forecasting methods using collections of well-studied time series datasets, such as the M3 data. The question is, though, how diverse and challenging are these time series, and do they enable us to study the unique strengths and weaknesses of different forecasting methods? This paper proposes a visualisation method for collections of time series that enables a time series to be represented as a point in a two-dimensional instance space. The effectiveness of different forecasting methods across this space is easy to visualise, and the diversity of the time series in an existing collection can be assessed. Noting that the diversity of the M3 dataset has been questioned, this paper also proposes a method for generating new time series with controllable characteristics in order to fill in and spread out the instance space, making our generalisations of forecasting method performances as robust as possible.

AB - It is common practice to evaluate the strength of forecasting methods using collections of well-studied time series datasets, such as the M3 data. The question is, though, how diverse and challenging are these time series, and do they enable us to study the unique strengths and weaknesses of different forecasting methods? This paper proposes a visualisation method for collections of time series that enables a time series to be represented as a point in a two-dimensional instance space. The effectiveness of different forecasting methods across this space is easy to visualise, and the diversity of the time series in an existing collection can be assessed. Noting that the diversity of the M3 dataset has been questioned, this paper also proposes a method for generating new time series with controllable characteristics in order to fill in and spread out the instance space, making our generalisations of forecasting method performances as robust as possible.

KW - Forecasting algorithm comparison

KW - M3-Competition

KW - Time series generation

KW - Time series visualisation

UR - http://www.scopus.com/inward/record.url?scp=85009115385&partnerID=8YFLogxK

U2 - 10.1016/j.ijforecast.2016.09.004

DO - 10.1016/j.ijforecast.2016.09.004

M3 - Article

VL - 33

SP - 345

EP - 358

JO - International Journal of Forecasting

JF - International Journal of Forecasting

SN - 0169-2070

IS - 2

ER -