I have a problem where I need to make predictions for a binary target $y$ given a set of features $X$ where $X$ is naturally nested in the form of repeated measurements. The data is meant to describe the behavior of people in a game where every line of data $X_t$ is timestamped and represent a behavior at time $t$. A toy example of my data can be the following :

ID      session      X1      X2      y
1          1         0.5     a       0
1          2         0.12    b       0
1          3         1.1     d       1
2          1         5.2     c       1
2          2         4.1     a       0

I have treated this as a typical classification problem with each sample $X_t$ being an $iid$ and an XGBoost classifier yields good enough results. However, many articles online state that this form of panel data (where we find repeated measurements per subject) is better handled with mixed effect models. This excellent blog post goes through mixed effect random forests (MERFs) for a regression problem but says that classification models are yet to be implemented (in $2017$ that is).

To this day, i know of no implementation for MERFs for classification problems. So my question is, did anyone come accross a similar classification task and solved it using mixed effect models? If yes is there any code implementation of such a model? (preferably python but I can do with R as well to test it).


1 Answer 1


I am actually looking for something similar, however, in my case I am looking for something to deal with multiclass labeled data.

In my search, I came across GPBoost, which may just satisfy your needs.


GPBoost should be both available for Python as well as R, and if I'm correct, binary classification is yet included.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.