We used three datasets when developing MIMS-unit algorithm. We list the datasets here you may refer to the datasets repos for details and use them to reproduce the results in the manuscript.

Elliptical shaker dataset

In this dataset, we collected elliptical shaker data from multiple devices at different spinning speed. Please refer to the manuscript for the experiment protocol.

Activpal 20 2
GT3X 30 3
GT3XPlus 40 6
GT3XPlus 80 6
GT9X 100 16
GT9X 60 8
LGUrbaneR 100 2
MotoGPlay 50 2

This dataset is used to

  1. Tune the best parameter setups for the extrapolation algorithm.
  2. Compute the cross device consistency vignette("cross_device_consistency") after applying different accelerometer summarization algorithms.

Dataset repo: https://github.com/mHealthGroup/MIMSunit-dataset-shaker
Dataset documentation: The file and folder names are self-explainable.

Human treadmill dataset

This dataset includes raw accelerometer data of treadmill walking and running from 10 participants wearing Actigraph devices (Some with GT3X+ and others with GT9X) on non dominant wrist and dominant waist.

The dataset is used to compare the consistency of MIMS-unit values and the other summarization algorithms (ENMO, Actigraph count) when people are walking and running at different speeds. The dataset is also used to generate the dominant frequency and amplitude figure of walking and running at different speeds in the supplementary materials.

Dataset repo: https://github.com/mHealthGroup/MIMSunit-dataset-treadmill
Dataset documentation: https://github.com/mHealthGroup/MIMSunit-dataset-treadmill

Human physical activity dataset (SPADES)

In this dataset, we collected 22 different daily physical activities from 50 participants with sensors mounted on seven different body locations (both wrists, both ankles, both waist sides, and dominant thigh).

Data from non dominant wrist and dominant ankle are used in developing MIMS-unit algorithm. It is mainly used to show the MIMS-unit values on different daily physical activities and specifically to show MIMS-unit algorithm does not suppress motions for sedentary activities.

Dataset repo: https://github.com/qutang/MUSS/releases/download/v1.2.0/muss_data.tar.gz
Dataset documentation: https://mhealth-specification.gitbook.io/mhealth-specification/

NHANES and NNYFS dataset

In addition to the datasets used to validate the algorithm, the MIMSunit algorithm was also used to process the NHANES dataset (2011-2014) and NNYFS dataset (2012).