OCLTS: One-Class Learning Time-Series Shapelets

Main Article Content

Akihiro Yamaguchi
Takeichiro Nishikawa

Abstract

Time-series shapelets are time-series segments effective for classifying time-series instances. In recent years, simultaneously learning both a classifier and the shapelets has been studied because they provide not only interpretable results but also superior classification performance. However, in some applications such as anomaly detection, class distributions are  highly imbalanced between majority and minority classes. In particular, it is important to detect unseen features, which do not appear during training, in a minority class if those features are discriminative. Our aim is to learn both a classifier and shapelets using only training instances for the majority class without the minority class. We propose a method called One-Class Learning Time-series Shapelets (OCLTS). OCLTS efficiently and simultaneously optimizes both the shapelets and a nonlinear classifier based on One-Class Support Vector Machine by a stochastic subgradient descent algorithm. Experimental results show the method’s effectiveness for interpretability and imbalanced binary classification.

Article Details

Section
Articles