Methods to estimate aspects of physical activity and sedentary behavior from high-frequency wrist accelerometer measurements

Abstract

This investigation developed models to estimate aspects of physical activity and sedentary behavior from three-axis high-frequency wrist-worn accelerometer data. The models were developed and tested on 20 participants (n = 10 males, n = 10 females, mean age = 24.1, mean body mass index = 23.9), who wore an ActiGraph GT3X+ accelerometer on their dominant wrist and an ActiGraph GT3X on the hip while performing a variety of scripted activities. Energy expenditure was concurrently measured by a portable indirect calorimetry system. Those calibration data were then used to develop and assess both machine-learning and simpler models with fewer unknown parameters (linear regression and decision trees) to estimate metabolic equivalent scores (METs) and to classify activity intensity, sedentary time, and locomotion time. The wrist models, applied to 15-s windows, estimated METs [random forest: root mean squared error (rSME) = 1.21 METs, hip: rMSE = 1.67 METs] and activity intensity (random forest: 75% correct, hip: 60% correct) better than a previously developed model that used counts per minute measured at the hip. In a separate set of comparisons, the simpler decision trees classified activity intensity (random forest: 75% correct, tree: 74% correct), sedentary time (random forest: 96% correct, decision tree: 97% correct), and locomotion time (random forest: 99% correct, decision tree: 96% correct) nearly as well or better than the machine-learning approaches. Preliminary investigation of the models' performance on two free-living people suggests that they may work well outside of controlled conditions.

Keywords: ActiGraph; GT3X+; high frequency; triaxial.

Copyright © 2015 the American Physiological Society.

Figures

Fig. 1.

A: relative performance of wrist methods and hip method to estimate metabolic equivalent scores (METs). “Hip LR” is the linear regression of Freedson et al. (10) and RC Hip 2-R is refined 2-regression method of Crouer et al. (6). B: relative performance of the wrist methods to classify MET level, sedentary time, and locomotion time. RF, random forest; NNET, neural network; SVM, support vector machine; LR, linear regression; CI, confidence interval.

Fig. 2.

A: residual plots of wrist and hip linear regression estimates of METs. Each point represents 15 s for the wrist linear regression, 10 s for the refined Crouter 2-regression method, and 60 s for the method of Freedson. B: residual plots of random forest, neural network, and support vector machine estimates of METs. Each point represents 15 s of data.

Fig. 3.

Mean performance of wrist models to estimate average METs per activity. TM, treadmill.

Fig. 4.

A: tree model to classify activity intensity. For A–C, each axis lists a summary statistic for the accelerometer signals. Those statistics are computed for each 15-s window that defines a point in on the plot. The region's label is the estimated class. B: tree model to classify sedentary time. C: tree model to classify locomotion time.

Fig. 5.

Evaluation of the methods vs. direct observation on 2 free-living subjects who are different from the subjects used to develop the models.