CLe Ver: A Feature Subset Selection Technique for Multivariate Time Series

Abstract

Feature subset selection (FSS) is one of the data pre-processing techniques to identify a subset of the original features from a given dataset before performing any data mining tasks. We propose a novel FSS method for Multivariate Time Series (MTS) based on Common Principal Components, termed CL e V er. It utilizes the properties of the principal components to retain the correlation information among original features while traditional FSS techniques, such as Recursive Feature Elimination (RFE), may lose it. In order to evaluate the effectiveness of our selected subset of features, classification is employed as the target data mining task. Our experiments show that CL e V er outperforms RFE and Fisher Criterion by up to a factor of two in terms of classification accuracy, while requiring up to 2 orders of magnitude less processing time.

This research has been funded in part by NSF grants EEC-9529152 (IMSC ERC), IIS-0238560 (PECASE) and IIS-0307908, and unrestricted cash gifts from Microsoft. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. The authors would like to thank Dr. Carolee Winstein and Jarugool Tretiluxana for providing us the BCAR dataset and valuable feedbacks, and Thomas Navin Lal for providing us the BCI MPI dataset. The authors would also like to thank the anonymous reviewers for their valuable comments.