Localized Sparse Principal Component Analysis of Multivariate Time Series in the Frequency Domain

Area of Science:

• Multivariate Analysis
• Time Series Analysis
• Frequency Domain Analysis

Background:

• Principal Component Analysis (PCA) is crucial for dimensionality reduction in multivariate data.
• Traditional PCA struggles with consistency and interpretability in high-dimensional settings.
• Interpretable PCA in time series requires sparse and frequency-localized principal components.

Purpose of the Study:

• To develop a consistent estimation procedure for interpretable Principal Component Analysis (PCA) in high-dimensional time series.
• To introduce a method for obtaining sparse and frequency-localized principal components.
• To apply the method to understand neurological mechanisms from EEG data.

Main Methods:

• Formulation of interpretable PCA for high-dimensional time series in the frequency domain.
• Development of a consistent estimation procedure.
• An efficient frequency-sequential algorithm for computing sparse-localized estimates.

Main Results:

• Consistent estimation of low-dimensional principal subspaces for high-dimensional time series.
• Sparse and frequency-localized principal component estimates.
• Demonstrated utility in analyzing high-density resting-state EEG data.

Conclusions:

• The proposed method enables consistent and interpretable PCA for high-dimensional time series.
• The frequency-sequential algorithm efficiently computes desired principal components.
• The approach offers insights into neurological mechanisms using EEG data.