Speaker: Jaroslaw (Jarek) Harezlak
From: Indiana University
Title: Raw accelerometry data – Structured Functional Principal Components to the rescue
Link: https://ncsu.zoom.us/j/94716351653

Abstract:

As walking is the most common form of physical activity, one’s physical health is likely related to walking features. However, current methods extracting walking features from raw accelerometry data do not account for varying cadence and commonality of features among subjects. We use data collected via ActiGraph GT3X+ devices in the Developmental Epidemiologic Cohort Study (DECOS) of older adults and apply structured functional principal component analysis (SFPCA) to extract features from walking signals on the subject-specific and the subject-spectrum level of a fast-paced 400m walk. We also use the subject level feature scores to study their associations with age and physical performance measures. Specifically, we transform raw data into the frequency domain by local Fast Fourier Transform to obtain walking spectra and use SFPCA to decompose these spectra into interpretable walking features. We found that 5 subject-specific and 19 subject-spectrum level features explained more than 85% of their respective level variation. We established that higher acceleration magnitude at the cadence was associated with younger age, lower BMI, faster average cadence and higher SPPB scores.