New model request: Mixed effect random forest (MERF)

[amcmahon17]

Hello, I have a new model request: Mixed effect random forest (MERF). There are a few packages that support them in R. I'm not sure if this is the proper package for it.

[hfrick]

Hi @amcmahon17 ! Could you say a bit more about your use case for them and point us to the most suitable implementations in R and references?

[amcmahon17]

Hello! Thank you for the response. Please let me know if the following is sufficient. The use case is repeated measures/longitudinal data and clustered data. It's basically the same as ordinary mixed effects. MERFs have been shown to have a performance boost vs ordinary random forests. The best package I have found is LongituRF. The relevant function is MERF(). I have references for two papers below. The second is by the author of the LongituRF package.

…

-https://cran.r-project.org/web/packages/LongituRF/index.html -"Mixed-effects random forest for clustered data" https://www.tandfonline.com/doi/abs/10.1080/00949655.2012.741599 (Journal of Statistical Computation and Simulation Volume 84, 2014 - Issue 6) -"Random forests for high-dimensional longitudinal data" https://journals.sagepub.com/doi/10.1177/0962280220946080 (Statistical Methods in Medical Research Volume 30, Issue 1) -"Comparing (Stochastic) Mixed-Effects Tree-Based Models in {LongituRF} & {ranger}", Blog of Ph.D. student Christopher Loan: https://www.christopherloan.com/blog/running-longitudinal-random-forests-with-longiturf/

On Mon, Dec 19, 2022 at 10:41 AM Hannah Frick ***@***.***> wrote: Hi @amcmahon17 <https://github.com/amcmahon17> ! Could you say a bit more about your use case for them and point us to the most suitable implementations in R and references? — Reply to this email directly, view it on GitHub <#50 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFVEQITDCWIIAR4SJDNV2ATWOB63RANCNFSM6AAAAAATCVE5NI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[amcmahon17]

Sorry, my references didn't come through for some reason. Here they are:

LongituRF package.

-https://cran.r-project.org/web/packages/LongituRF/index.html
-"Mixed-effects random forest for clustered data"
https://www.tandfonline.com/doi/abs/10.1080/00949655.2012.741599
 (Journal of Statistical Computation and Simulation
Volume 84, 2014 - Issue 6)

-"Random forests for high-dimensional longitudinal data"
https://journals.sagepub.com/doi/10.1177/0962280220946080
(Statistical Methods in Medical Research Volume 30, Issue 1)

-"Comparing (Stochastic) Mixed-Effects Tree-Based Models in {LongituRF} & {ranger}", Blog of Ph.D. student Christopher Loan: https://www.christopherloan.com/blog/running-longitudinal-random-forests-with-longiturf/

[hfrick]

Thank you! I'll give this a closer look the next time we are returning to more multilevelmod development 🪆

[jscamac]

This is a really nicely written paper that summarises the use-case and some of the R packages available in R.
https://academic.oup.com/bib/article/24/2/bbad002/6991123

[marianawebb]

Agreed that it would be awesome to have MERF (or something similar like MixRF) in tidymodels. I'll be following

[sebsilas]

Anyone been working on this? It would be great to have!

[jprice80]

I agree that this would be great to add. Longitudinal use cases are increasingly more common. Consequently, this would be great to add as less intensive alternatives compared to LSTM keras models.

[UrielMenalled]

Agreed! MERFs would be great to have in TidyModels!