Learning Method for Time-Series Shapelet Evolution

DOI

Bibliographic Information

Other Title
  • 変形可能な判別波形パターンの学習法

Abstract

With the spread of IoT, the need for time-series classification using machine learning is increasing in industrial fields such as infrastructure, medicine, and manufacturing. In recent years, methods that jointly learn classifiers and discriminative subsequences called shapelets have attracted attention due to their interpretability and superior classification performance. In this paper, we propose the concept of an evolvable shapelet, whose shape changes with seasonality, human habituation, and machine degradation, and demonstrate their effectiveness in each industrial field. The proposed method jointly learns not only shapelets and a classifier but also regression models for predicting shapelet evolution.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top