Self-labeling techniques for semi-supervised time series classification: an empirical study

Mabel González, Christoph Bergmeir, Isaac Triguero, Yanet Rodríguez, José M. Benítez

    Research output: Contribution to journalArticleResearchpeer-review

    Abstract

    An increasing amount of unlabeled time series data available render the semi-supervised paradigm a suitable approach to tackle classification problems with a reduced quantity of labeled data. Self-labeled techniques stand out from semi-supervised classification methods due to their simplicity and the lack of strong assumptions about the distribution of the labeled and unlabeled data. This paper addresses the relevance of these techniques in the time series classification context by means of an empirical study that compares successful self-labeled methods in conjunction with various learning schemes and dissimilarity measures. Our experiments involve 35 time series datasets with different ratios of labeled data, aiming to measure the transductive and inductive classification capabilities of the self-labeled methods studied. The results show that the nearest-neighbor rule is a robust choice for the base classifier. In addition, the amending and multi-classifier self-labeled-based approaches reveal a promising attempt to perform semi-supervised classification in the time series context.

    Original languageEnglish
    Pages (from-to)493-528
    Number of pages36
    JournalKnowledge and Information Systems
    Volume55
    Issue number2
    DOIs
    Publication statusPublished - May 2018

    Keywords

    • Self-labeled
    • Self-training
    • Semi-supervised classification
    • Semi-supervised learning
    • Time series classification

    Cite this

    González, Mabel ; Bergmeir, Christoph ; Triguero, Isaac ; Rodríguez, Yanet ; Benítez, José M. / Self-labeling techniques for semi-supervised time series classification : an empirical study. In: Knowledge and Information Systems. 2018 ; Vol. 55, No. 2. pp. 493-528.
    @article{c9bb16e9ada541cc800637642f59f953,
    title = "Self-labeling techniques for semi-supervised time series classification: an empirical study",
    abstract = "An increasing amount of unlabeled time series data available render the semi-supervised paradigm a suitable approach to tackle classification problems with a reduced quantity of labeled data. Self-labeled techniques stand out from semi-supervised classification methods due to their simplicity and the lack of strong assumptions about the distribution of the labeled and unlabeled data. This paper addresses the relevance of these techniques in the time series classification context by means of an empirical study that compares successful self-labeled methods in conjunction with various learning schemes and dissimilarity measures. Our experiments involve 35 time series datasets with different ratios of labeled data, aiming to measure the transductive and inductive classification capabilities of the self-labeled methods studied. The results show that the nearest-neighbor rule is a robust choice for the base classifier. In addition, the amending and multi-classifier self-labeled-based approaches reveal a promising attempt to perform semi-supervised classification in the time series context.",
    keywords = "Self-labeled, Self-training, Semi-supervised classification, Semi-supervised learning, Time series classification",
    author = "Mabel Gonz{\'a}lez and Christoph Bergmeir and Isaac Triguero and Yanet Rodr{\'i}guez and Ben{\'i}tez, {Jos{\'e} M.}",
    year = "2018",
    month = "5",
    doi = "10.1007/s10115-017-1090-9",
    language = "English",
    volume = "55",
    pages = "493--528",
    journal = "Knowledge and Information Systems",
    issn = "0219-1377",
    publisher = "Springer-Verlag London Ltd.",
    number = "2",

    }

    Self-labeling techniques for semi-supervised time series classification : an empirical study. / González, Mabel; Bergmeir, Christoph; Triguero, Isaac; Rodríguez, Yanet; Benítez, José M.

    In: Knowledge and Information Systems, Vol. 55, No. 2, 05.2018, p. 493-528.

    Research output: Contribution to journalArticleResearchpeer-review

    TY - JOUR

    T1 - Self-labeling techniques for semi-supervised time series classification

    T2 - an empirical study

    AU - González, Mabel

    AU - Bergmeir, Christoph

    AU - Triguero, Isaac

    AU - Rodríguez, Yanet

    AU - Benítez, José M.

    PY - 2018/5

    Y1 - 2018/5

    N2 - An increasing amount of unlabeled time series data available render the semi-supervised paradigm a suitable approach to tackle classification problems with a reduced quantity of labeled data. Self-labeled techniques stand out from semi-supervised classification methods due to their simplicity and the lack of strong assumptions about the distribution of the labeled and unlabeled data. This paper addresses the relevance of these techniques in the time series classification context by means of an empirical study that compares successful self-labeled methods in conjunction with various learning schemes and dissimilarity measures. Our experiments involve 35 time series datasets with different ratios of labeled data, aiming to measure the transductive and inductive classification capabilities of the self-labeled methods studied. The results show that the nearest-neighbor rule is a robust choice for the base classifier. In addition, the amending and multi-classifier self-labeled-based approaches reveal a promising attempt to perform semi-supervised classification in the time series context.

    AB - An increasing amount of unlabeled time series data available render the semi-supervised paradigm a suitable approach to tackle classification problems with a reduced quantity of labeled data. Self-labeled techniques stand out from semi-supervised classification methods due to their simplicity and the lack of strong assumptions about the distribution of the labeled and unlabeled data. This paper addresses the relevance of these techniques in the time series classification context by means of an empirical study that compares successful self-labeled methods in conjunction with various learning schemes and dissimilarity measures. Our experiments involve 35 time series datasets with different ratios of labeled data, aiming to measure the transductive and inductive classification capabilities of the self-labeled methods studied. The results show that the nearest-neighbor rule is a robust choice for the base classifier. In addition, the amending and multi-classifier self-labeled-based approaches reveal a promising attempt to perform semi-supervised classification in the time series context.

    KW - Self-labeled

    KW - Self-training

    KW - Semi-supervised classification

    KW - Semi-supervised learning

    KW - Time series classification

    UR - http://www.scopus.com/inward/record.url?scp=85027066432&partnerID=8YFLogxK

    U2 - 10.1007/s10115-017-1090-9

    DO - 10.1007/s10115-017-1090-9

    M3 - Article

    VL - 55

    SP - 493

    EP - 528

    JO - Knowledge and Information Systems

    JF - Knowledge and Information Systems

    SN - 0219-1377

    IS - 2

    ER -