Description

Two “Sojuourn” models were developed for estimating energy expenditure in adults using either uniaxial or triaxial counts from an ActiGraph worn on the right hip. Models, which are a combination of decision tree and artificial neural network, were developed using free-living observation and cross-validated in a hold-out sample.

The following two models are available. Root Mean Square Error (RMSE) is also provided.

• Vertical axis counts/s model (1 MET-hr)

• Counts/s from three axes model (1.0 MET-hr)

Development/Validation

Sample: 13 (5 males) healthy adults, 24.8±5.9 years of age

Setting: Free-living

Activities: Free-living

Criterion: Direct observation and compendium

Accelerometer(s): ActiGraph GT3X on right hip

Validation approach: Holdout of 7 participants

Phase Designation

(What's this?)

This model is in Phase 3.

Instructions

The R code from the original paper is provided below. However, it was recently discovered that this code contains an error (described here). This error has been fixed when the model is implemented using the Sojourn R package. While we are retaining the below code for posterity, we recommend using the Sojourn R package to implement the correct/fixed version.

The R code can be used to read in the necessary models and functions and a csv data file, then use the Sojourn model to predict energy expenditure for each 1-s window. There is an example data sheet which is a csv file in the format exported by ActiGraph (10 header rows followed by counts in each axis). Further instructions are in the R code. More information about R is found here.

This code relies on R software which can be downloaded for free at https://www.r-project.org/

Attached Files

Download all files as a .zip or download individual files below.

Code

Example data

Necessary R models (nnet3ests, cent.1, scal.1, class.nnn.use.this) and functions

Note: Compared to the original source, code was edited for consistency but models/functions and example data were unchanged

Source Information

Reference:

Lyden, K., Keadle, S. K., Staudenmayer, J., & Freedson, P. S. (2014). A method to estimate free-living active and sedentary behavior from an accelerometer. Medicine and Science in Sports and Exercise, 46(2), 386.

https://www.doi.org/10.1249/MSS.0b013e3182a42a2d Link to paper

Original source for code:

www.math.umass.edu/~jstauden/SojournCode.zip

Corresponding author: Patty Freedson, psf@kin.umass.edu

Repository contact: Kimberly A. Clevenger, clevengerkimberly@gmail.com

Contact

Kimberly Clevenger at accelerometerrepository@gmail.com