Off-campus UMass Amherst users: To download dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users, please click the view more button below to purchase a copy of this dissertation from Proquest.

(Some titles may also be available free of charge in our Open Access Dissertation Collection, so please check there first.)

Refinement, validation and application of a machine learning method for estimating physical activity and sedentary behavior in free-living people

Kate Lyden, University of Massachusetts Amherst

Abstract

There is limited knowledge of the dose-response relationship between physical activity (PA), sedentary behavior (SB) and health. Poor measures of free-living PA and SB exposure are major contributing factors to these knowledge gaps. The overall objective of this dissertation was to address these issues by refining, validating and applying a machine-learning methodology for measuring PA and SB for use in free-living people. By combining neural networks and decision tree analyses we developed a method better suited for use in free-living people. Our new method is called the sojourn method and it estimates PA and SB from a single hip mounted accelerometer. ^ Study 1 validated two versions of this method: sojourn-1x (soj-1x) and sojourn-3x (soj-3x). Soj-1x uses data from a vertical accelerometer sensor, while soj-3x uses r data from the vertical, anterior-posterior and medial-lateral accelerometer sensors. Seven participants were directly observed in the free-living environment for ten consecutive hours on three separate occasions. PA and SB estimated from soj-1x, soj-3x and a neural network previously calibrated in the laboratory (lab-nnet) were compared to direct observation. Compared to the lab-nnet, soj-1x and soj-3x improved estimates of MET-hours (lab-nnet: bias (95% CI) = 5.4 (4.6-6.2), rMSE = 5.4 (4.6-6.2), soj-1x: bias = 0.3 (-0.2-0.9), rMSE = 1.0 (0.6-1.3), soj-3x: bias = 0.5 (-0.1-1.1), rMSE = 1.1 (0.7-1.5)) and minutes in different intensity categories (lab-nnet: rMSE range = 10.2 (vigorous) - 55.0 (light), soj-1x: rMSE range = 4.0 (MVPA) - 50.1 (sedentary), soj-3x: rMSE range = 7.8 (MVPA) - 27.8 (light)). Soj-1x and soj-3x also produced accurate estimates of qualifying minutes, qualifying bouts, breaks from sedentary time and break-rate. ^ Study 2 evaluated the sensitivity of soj-1x and soj-3x to detect change in habitual activity. Thirteen participants completed three, seven day conditions: sedentary, moderately active and very active. Soj-1x and soj-3x were sensitive to change in MET-hours (mean (95% CI): soj-1x: sedentary = 19.8 (19.0-20.7), moderately active = 22.7 (22.0-23.4), very active = 27.0 (25.8-28.2), soj-3x: sedentary = 18.2 (17.7-18.8), moderately active = 22.3 (21.6-23.1), very active = 27.6 (26.4-28.7)) and time in different intensity categories. ^ Study 3 applied soj-3x to a free-living intervention to elucidate the effects of increased sedentary behavior on markers of cardiometabolic health. Eleven participants completed seven days of an active condition followed by seven days of an inactive condition. Insulin action significantly decreased 17% (5.4-30.2), while total cholesterol, LDL and HDL did not change from the active to inactive condition. This dissertation used novel methods to improve PA and SB estimation in a free-living environment and to improve our understanding of the physiologic response to increased free-living SB. These methods ultimately have the potential to broaden our understanding of how PA and SB dose are linked to health.^

Subject Area

Kinesiology

Recommended Citation

Lyden, Kate, "Refinement, validation and application of a machine learning method for estimating physical activity and sedentary behavior in free-living people" (2012). Doctoral Dissertations Available from Proquest. AAI3545961.
http://scholarworks.umass.edu/dissertations/AAI3545961

Share

COinS