Time series feature learning with labeled and unlabeled data

作者：

Highlights：

• A novel time series feature selection task with labeled and unlabeled data.

• A new semi-supervised time series feature learning model is proposed.

• The model integrates least square minimization, spectral analysis, scaled pseudo labels as well as time series feature similarity regularization terms.

• Experiments on real-world data demonstrating significant performance gain of the proposed model.

摘要

•A novel time series feature selection task with labeled and unlabeled data.•A new semi-supervised time series feature learning model is proposed.•The model integrates least square minimization, spectral analysis, scaled pseudo labels as well as time series feature similarity regularization terms.•Experiments on real-world data demonstrating significant performance gain of the proposed model.

论文关键词：Time series,Feature selection,Semi-supervised learning,Classification

论文评审过程：Received 28 September 2017, Revised 6 December 2018, Accepted 18 December 2018, Available online 19 December 2018, Version of Record 2 January 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.12.026