OxWearables/biobankAccelerometerAnalysis

Activity classification model performances and sampling rate

Closed this issue · 4 comments

ghammad commented

If I am not mistaken, the default sampling rate for the data used to develop and validate the model described in Willetts et al., 2018 and made available through the "biobankAcc" package, is 100Hz.

Does the impact of the sampling rate on the model performances have been studied? I.e. is it possible to use the model with data acquired at a lower sampling rate (ex. 50 Hz)?

If I understand correctly, features in a frequency domain up to 15Hz were used... so still below the Nyquist frequency for a sampling rate of 50Hz... so there is hope, isn't it?

Thank you.

chanshing commented

Hi. Yes, you can use a lower sampling rate. I think our features are resolution-invariant so it should work.

ghammad commented

Ok. Thank you.

  • 👍1
aidendoherty commented

Thanks for this question @ghammad
This paper might also be of interest:

Small, S., Khalid, S., Dhiman, P., Chan, S., Jackson, D., Doherty, A., & Price, A. (2021). Impact of Reduced Sampling Rate on Accelerometer-Based Physical Activity Monitoring and Machine Learning Activity Classification, Journal for the Measurement of Physical Behaviour, 4(4), 298-310. Retrieved Dec 3, 2021, from https://journals.humankinetics.com/view/journals/jmpb/4/4/article-p298.xml

ghammad commented

Hi @aidendoherty

Thank you for the reference. I missed it! Very useful paper as, working in a chronobiology lab, from our perspective, the added value of actigraphy with respect to other techniques (among other advantages), resides in long recordings.

Cheers,

Greg