Description

Six artificial neural networks were developed for estimating energy expenditure in adults using a GENEActiv worn on either wrist and three different raw acceleration feature sets. Adults performed simulated activities of daily living, locomotion, and other exercises and the model was cross-validated using leave-one-out cross validation. This paper also developed algorithms for predicting on which wrist a monitor was worn but these are not included in the repository at this time.

The following models are available. Root Mean Square Error (RMSE) for individual models was not provided but overall, was 1.47 METs for feature set 1, 1.39 METs for feature set 2, and 1.25 METs for feature set 3.

• Artificial neural network for left wrist monitor using feature set 1

• Artificial neural network for left wrist monitor using feature set 2

• Artificial neural network for left wrist monitor using feature set 3

• Artificial neural network for right wrist monitor using feature set 1

• Artificial neural network for right wrist monitor using feature set 2

• Artificial neural network for right wrist monitor using feature set 3

Development/Validation

Sample: 39 (19 males) healthy adults, 18-44 years of age

Setting: Laboratory

Activities: Activities of daily living, cycling, walking and running, resistance exercise, stairs, stationary behaviors

Criterion: Oxycon Mobile (VO₂)

Accelerometer(s): GENEActiv on either wrist

Validation approach: Leave-one-out cross-validation

Phase Designation

(What's this?)

This model is in Phase 0 and 2.

Instructions

The R code can be used to read in a csv file with the necessary features and then use the neural network to predict energy expenditure for each 30-s window. There is a separate file for each of the six models. There is an example data sheet for each of the monitor wear locations (left and right wrist) with the variables described in the data dictionary that need to be present in the csv file. 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 left wrist data

Example right wrist data

Data dictionary

Left wrist model feature set 1

Left wrist model feature set 2

Left wrist model feature set 3

Right wrist model feature set 1

Right wrist model feature set 2

Right wrist model feature set 3

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

Source Information

Reference:

Montoye, A. H., Pivarnik, J. M., Mudd, L. M., Biswas, S., & Pfeiffer, K. A. (2016). Wrist-independent energy expenditure prediction models from raw accelerometer data. Physiological Measurement, 37(10), 1770. https://doi.org/10.1088/0967-3334/37/10/1770 Link to paper

Original source for code:

https://drive.google.com/open?id=0B-BgdTzyd2OxRDZTQWVxZW1qMzA

Corresponding author: Alexander H.K. Montoye, montoyeah@alma.edu

Contact

Kimberly Clevenger at accelerometerrepository@gmail.com