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 https://github.com/tidymodels/multilevelmod/issues/50#issuecomment-1357861907, 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