Extracting Actigraphy-Based Walking Features with Structured Functional Principal Components

• Activity Number: 384 - Next-Generation Sequencing and High-Dimensional Data
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
• Sponsor: Biometrics Section
• Author(s): Verena Werkmann* and Jaroslaw Harezlak and Nancy W. Glynn
• Companies: Indiana University and Indiana University and University of Pittsburgh
• Keywords: accelerometry time series data; functional principal components; Fourier transform; walking cadence

Abstract

Current methods applied to walking feature extraction from raw accelerometry data do not take into account varying cadence and information shared among subjects. We apply structured functional principal component analysis (SFPCA) to extract features from walking spectra on both population and intra-subject level. Also, we use scores of the population level features to study their associations with health indicators. We use raw accelerometry data collected on 48 elderly individuals as part of the Developmental Epidemiologic Cohort Study (DECOS). To obtain walking spectra, we transform the raw accelerometry data into the frequency domain by applying local Fast Fourier Transform. SFPCA decomposes the spectra into easily interpretable 7 population level and 39 intra-subject level features expressed in terms of cadence and acceleration. Our results show that 68% of the total data variation arises at the population level while the intra-subject share is 32%. Moreover, we find that lower acceleration magnitude at the cadence is associated with lower gait speeds and an older age. Higher acceleration magnitude at cadence multiples of 2.5 and 3.5 are only related to higher gait speeds.