Including Temperature Data for Non-Wear Detection in GGIR

[Hananeh Younesian]

Hi Vincent hi all,

I hope you are doing great.

In my study, participants wore an AX6 sensor on their lower back continuously for 24 hours over 7 days. I ran GGIR on the raw .csv data two times:

(1) using only accelerations data (rmc.col.acc argument) 

(2) using accelerations and body temprature (rmc.col.acc & rmc.col.temp arguments)

In both cases, the number of valid hours was the same.

My questions are: How can I effectively include temperature data in GGIR for non-wear detection? Is there a specific setting or argument I need to adjust to ensure temperature is used in the non-wear algorithm?

Thanks in advance for your help.

Kind regards,
Hana

[Vincent van Hees]

Hi Hana,

GGIR does not offer such functionality.

Do you know any specific algorithm GGIR should be using to account for temperature in the non-wear detection?

Kind regards,

Vincent

--
You received this message because you are subscribed to the Google Groups "R package GGIR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to RpackageGGIR...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/RpackageGGIR/6abe41d1-2347-4158-869b-e4bb0286a29cn%40googlegroups.com.

[Zachary Hubshman]

Hi,

I came across an algorithm that attempts to detect non-wear by combining your approach with a temperature-based detection method. However, I noticed that the implementation is based on your 2011 paper, which is somewhat outdated compared to your later work—particularly the refinements described in your 2013 paper and the more recent implementation in GGIR (as of 2023).

Reference:
https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-022-01633-6

I'm currently experimenting with some Python code to see whether incorporating the updates from your later work improves performance. To be honest, I suspect this will mainly optimize edge cases—such as detecting the start and end of non-wear periods more precisely—rather than dramatically improving overall non-wear detection. So might be nice in refining it, I can let you know what I end up finding out once I'm done working on it.

Best,
Zachary

[Zachary Hubshman]

Clarification,

Actually it's only loosely based on the 2013 algorithm. Seems more like it used the 2013 algorithm as a comparison point but ended up getting a similar way to detect non-wear (two axes below certain threshold). But there may be useful information to glean from the paper nonetheless.

Best,

Zachary

[Vincent van Hees]

Thanks Zachary,

I agree that there is probably no major benefit in using temperature for most cases.

I did not use temperature all those years because I was concerned that it would actually introduce error given that temperature sensor values are influenced by many factors other than wearing the accelerometer or not, e.g. clothing, environmental temperature, tightness of the strap,  and accuracy of the calibration of the temperature sensor.

It would require a rather sophisticated study to demonstrate the added value of temperature despite variations in all those factors.

Best,

Vincent

To view this discussion, visit https://groups.google.com/d/msgid/RpackageGGIR/218498d2-9e58-47a0-96b4-fdc93347f563n%40googlegroups.com.