My data consists of presence/absence (PA) of a trait in 354 plants collected from 127 collection sites as response, and a set of 25 climatic continuous variables in each site as predictors. The sample size for each collection site ranges from 1 to 4 plants. My goal is to find the climatic variables that are the best predictors of the PA pattern. As expected, the climate data is highly correlated. I would like to use the random forest approach to find the variables.

I have 3 questions:

  1. In gee-glm (generalized estimating equations) one can use a mixed model with the collection site as clustering variable (random effect) . Is there a way or a need to define such random effect in random forest analysis or I can just add the collection site as a factor to the predictors?
  2. Alternatively, I can calculate the frequency of the trait for each collection site and use it as the response (each case is a collection site). Then I will not need the collection site as random effect. However, the sample size is very small 1-4 plants. Is this a better approach than doing the analysis as logistic regression where each case is an individual?
  3. Do I need to select the most uncorrelated variables? or random forest can handle many correlated variables?

Thank you.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.