Discrepancies in NHANES Step Count Results

[j262byuu]

Hi,

First of all, thank you for creating and maintaining the excellent OxWearables/stepcount package!

I am not a participant in the 2024 Reproducible Machine Learning of Wearables in Health Data Science course, but I am conducting similar research—classifying step counts in the NHANES 2011–2014 cycle.

I used exactly the same method described in 01-nhanes_data_exploration.Rmd, which involves downloading raw 80Hz data, appending them, and then passing the massive csv file to OxWearables/stepcount (I'm using version 3.9.0 with SSL). However, when I compared my results to the output in nhanes_stepcount.csv, I noticed some discrepancies. Specifically, while WearTime(days) and NonwearTime(days) match, all other results differ between the two results (see below, my results on the left and those from nhanes_stepcount.csv on the right).

Could these differences be attributed to variations in the model settings or the version of OxWearables/stepcount? I’d appreciate any insights or guidance you can provide.

Thanks,
Xiaoyu