Description

Twenty-four models were developed for predicting energy expenditure in young children using accleration metrics from one of six monitor/wear location combinations (ActiGraph at each hip, GENEActiv at each wrist and the right hip, ActivPal on right thigh). Models were developed from children performing indoor and outdoor age-appropriate activities using leave-one-out cross-validation. Another 48 models are available for kJ/min and J/min/kg but are not included in this repository.

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

• Artificial neural network for ActiGraph on right hip (1.66 METs)

• Linear model for ActiGraph on right hip (1.70 METs)

• Mixed linear model for ActiGraph on right hip (1.70 METs)

• Random forest for ActiGraph on right hip (1.56 METs)

• Artificial neural network for ActiGraph on left hip (1.63 METs)

• Linear model for ActiGraph on left hip (1.67 METs)

• Mixed linear model for ActiGraph on left hip (1.67 METs)

• Random forest for ActiGraph on left hip (1.52 METs)

• Artificial neural network for GENEActiv on right hip (1.66 METs)

• Linear model for GENEActiv on right hip (1.68 METs)

• Mixed linear model for GENEActiv on right hip (1.69 METs)

• Random forest for GENEActiv on right hip (1.53 METs)

• Artificial neural network for GENEActiv on right wrist (1.62 METs)

• Linear model for GENEActiv on right wrist (1.69 METs)

• Mixed linear model for GENEActiv on right wrist (1.69 METs)

• Random forest for GENEActiv on right wrist (1.48 METs)

• Artificial neural network for GENEActiv on left wrist (1.61 METs)

• Linear model for GENEActiv on left wrist (1.70 METs)

• Mixed linear model for GENEActiv on left wrist (1.69 METs)

• Random forest for GENEActiv on left wrist (1.47 METs)

• Artificial neural network for ActivPal on right thigh (1.81 METs)

• Linear model for ActivPal on right thigh (1.69 METs)

• Mixed linear model for ActivPal on right thigh (1.69 METs)

• Random forest for ActivPal on right thigh (1.56 METs)

Development/Validation

Sample: 41 (22 males) healthy children, 3.0-6.3 years of age

Setting: School

Activities: Free-choice, games, overground/self-paced walking and running, stationary

Criterion: MetaMax 3b (VO₂)

Accelerometer(s): ActiGraph right hip, ActiGraph left hip, GENEActiv right hip, GENEActiv right wrist, GENEActiv left wrist, ActivPal right thigh (models are for one location/monitor so not all needed)

Validation approach: Leave-one-out cross-validation

Phase Designation

(What's this?)

This model is in Phase 0, 2, and 4.

Instructions

The R code can be used to read in a csv file with the necessary variables (as described in the data dictionary and seen in the sample data). The chosen model can then be used to predict energy expenditure. The authors also provide some code for extracting features. 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
Data dictionary

Sample data

Data conversion code (not edited/used by repository)

ActiGraph artificial neural network right hip

ActiGraph Linear model right hip

ActiGraph Mixed linear model right hip

ActiGraph Random forest for ActiGraph right hip

ActiGraph Artificial neural network for ActiGraph left hip

ActiGraph Linear model for ActiGraph left hip

ActiGraph Mixed linear model for ActiGraph left hip

ActiGraph Random forest for ActiGraph left hip

GENEActiv Artificial neural network for GENEActiv right hip

GENEActiv Linear model for GENEActiv right hip

GENEActiv Mixed linear model for GENEActiv right hip

GENEActiv Random forest for GENEActiv right hip

GENEActiv Artificial neural network for GENEActiv right wrist

GENEActiv Linear model for GENEActiv right wrist

GENEActiv Mixed linear model for GENEActiv right wrist

GENEActiv Random forest for GENEActiv right wrist

GENEActiv Artificial neural network for GENEActiv left wrist

GENEActiv Linear model for GENEActiv left wrist

GENEActiv Mixed linear model for GENEActiv left wrist

GENEActiv Random forest for GENEActiv left wrist

ActivPal Artificial neural network for ActivPal right thigh

ActivPal Linear model for ActivPal right thigh

ActivPal Mixed linear model for ActivPal right thigh

ActivPal Random forest for ActivPal right thigh

Note: Compared to the original source, code was edited for consistency, data dictionary was added, models were unchanged

Source Information

Reference:

Steenbock, B., Wright, M. N., Wirsik, N., & Brandes, M. (2019). Accelerometry-based prediction of energy expenditure in preschoolers. Journal for the Measurement of Physical Behaviour, 2(2), 94-102. https://doi.org/10.1123/jmpb.2018-0032 Link to article

Original source for code:

https://github.com/bips-hb/EE_prediction

Corresponding author: Berit Steenbock, steenbock@leibniz-bips.de

Contact

Kimberly Clevenger at accelerometerrepository@gmail.com