Clustering of interval time series

Elizabeth Ann Maharaj, Paulo Teles, Paula Brito

Research output: Contribution to journalArticleResearchpeer-review

Abstract

Interval time series occur when real intervals of some variable of interest are registered as an ordered sequence along time. We address the problem of clustering interval time series (ITS), for which different approaches are proposed. First, clustering is performed based on point-to-point comparisons. Time-domain and wavelet features also serve as clustering variables in alternative approaches. Furthermore, autocorrelation matrix functions, gathering the autocorrelation and cross-correlation functions of the ITS upper and lower bounds, may be compared using adequate distances (e.g. the Frobenius distance) and used for clustering ITS. An improved procedure to determine the autocorrelation function of ITS is proposed, which also serves as a basis for clustering. The different alternative approaches are explored and their performances compared for ITS simulated under different setups. An application to sea level daily ranges, observed at different locations in Australia, illustrates the proposed methods.

Original languageEnglish
Number of pages24
JournalStatistics and Computing
DOIs
Publication statusAccepted/In press - 21 Jan 2019

Keywords

  • Interval autocorrelation
  • Interval data
  • Interval time series
  • Time series clustering

Cite this

@article{0b39dd9bd6164815829aca58e0647705,
title = "Clustering of interval time series",
abstract = "Interval time series occur when real intervals of some variable of interest are registered as an ordered sequence along time. We address the problem of clustering interval time series (ITS), for which different approaches are proposed. First, clustering is performed based on point-to-point comparisons. Time-domain and wavelet features also serve as clustering variables in alternative approaches. Furthermore, autocorrelation matrix functions, gathering the autocorrelation and cross-correlation functions of the ITS upper and lower bounds, may be compared using adequate distances (e.g. the Frobenius distance) and used for clustering ITS. An improved procedure to determine the autocorrelation function of ITS is proposed, which also serves as a basis for clustering. The different alternative approaches are explored and their performances compared for ITS simulated under different setups. An application to sea level daily ranges, observed at different locations in Australia, illustrates the proposed methods.",
keywords = "Interval autocorrelation, Interval data, Interval time series, Time series clustering",
author = "Maharaj, {Elizabeth Ann} and Paulo Teles and Paula Brito",
year = "2019",
month = "1",
day = "21",
doi = "10.1007/s11222-018-09851-z",
language = "English",
journal = "Statistics and Computing",
issn = "0960-3174",
publisher = "Springer",

}

Clustering of interval time series. / Maharaj, Elizabeth Ann; Teles, Paulo; Brito, Paula.

In: Statistics and Computing, 21.01.2019.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Clustering of interval time series

AU - Maharaj, Elizabeth Ann

AU - Teles, Paulo

AU - Brito, Paula

PY - 2019/1/21

Y1 - 2019/1/21

N2 - Interval time series occur when real intervals of some variable of interest are registered as an ordered sequence along time. We address the problem of clustering interval time series (ITS), for which different approaches are proposed. First, clustering is performed based on point-to-point comparisons. Time-domain and wavelet features also serve as clustering variables in alternative approaches. Furthermore, autocorrelation matrix functions, gathering the autocorrelation and cross-correlation functions of the ITS upper and lower bounds, may be compared using adequate distances (e.g. the Frobenius distance) and used for clustering ITS. An improved procedure to determine the autocorrelation function of ITS is proposed, which also serves as a basis for clustering. The different alternative approaches are explored and their performances compared for ITS simulated under different setups. An application to sea level daily ranges, observed at different locations in Australia, illustrates the proposed methods.

AB - Interval time series occur when real intervals of some variable of interest are registered as an ordered sequence along time. We address the problem of clustering interval time series (ITS), for which different approaches are proposed. First, clustering is performed based on point-to-point comparisons. Time-domain and wavelet features also serve as clustering variables in alternative approaches. Furthermore, autocorrelation matrix functions, gathering the autocorrelation and cross-correlation functions of the ITS upper and lower bounds, may be compared using adequate distances (e.g. the Frobenius distance) and used for clustering ITS. An improved procedure to determine the autocorrelation function of ITS is proposed, which also serves as a basis for clustering. The different alternative approaches are explored and their performances compared for ITS simulated under different setups. An application to sea level daily ranges, observed at different locations in Australia, illustrates the proposed methods.

KW - Interval autocorrelation

KW - Interval data

KW - Interval time series

KW - Time series clustering

UR - http://www.scopus.com/inward/record.url?scp=85060555984&partnerID=8YFLogxK

UR - https://monash.figshare.com/s/3f8da9957905ada3864d

U2 - 10.1007/s11222-018-09851-z

DO - 10.1007/s11222-018-09851-z

M3 - Article

JO - Statistics and Computing

JF - Statistics and Computing

SN - 0960-3174

ER -