Scholarlink

A large volume ofThunderstorm dataClassification related to wind has been recorded over the last decade in Mediterranean ports in order to better understand the effects of thunderstorm windsThunderstorm winds on civil engineering structures. Automated classificationClassification techniques have thus been developed to detect these events of interest in such large databases. To ensure the autonomy and interpretability of the process, it is convenient to use a machine learningMachine learning classifier trained on shapelet transformsShapelet transform. Techniques such as randomized sampling method, rotation forest classifier and ensemble voting rule are utilized in this paper to accelerate the discovery of shapelets while continuing to increase the accuracy and the stability of the procedure. These improvements in terms of both computational and operational efficiency regarding the identification of thunderstormsThunderstorm are assessed in this paper. Overall, the time needed to discover shapelets is divided by 10–100 when the same ratio of candidates is selected from the entire pool. Doing so does not harm the classificationClassification process afterwards, even though the best shapelets are slightly less discriminative.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improving the Discovery of Shapelets in Time Series for Thunderstorm Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Margaux Geuzaine,
  • Monica Arul,
  • Ahsan Kareem

摘要

A large volume ofThunderstorm dataClassification related to wind has been recorded over the last decade in Mediterranean ports in order to better understand the effects of thunderstorm windsThunderstorm winds on civil engineering structures. Automated classificationClassification techniques have thus been developed to detect these events of interest in such large databases. To ensure the autonomy and interpretability of the process, it is convenient to use a machine learningMachine learning classifier trained on shapelet transformsShapelet transform. Techniques such as randomized sampling method, rotation forest classifier and ensemble voting rule are utilized in this paper to accelerate the discovery of shapelets while continuing to increase the accuracy and the stability of the procedure. These improvements in terms of both computational and operational efficiency regarding the identification of thunderstormsThunderstorm are assessed in this paper. Overall, the time needed to discover shapelets is divided by 10–100 when the same ratio of candidates is selected from the entire pool. Doing so does not harm the classificationClassification process afterwards, even though the best shapelets are slightly less discriminative.