I'm interested in use conditional random forest trees for variable selection, in order to reduce bias when correlated or heterogeneous-format variables are used. In R it is possible, but I'm looking for the same in Knime.
unfortunately there is currently no native KNIME node that provides the functionality of conditional random forests.
But you can easily use the R nodes to perform tasks in R from within your KNIME workflow. The R package that you would want to use is called party and the corresponding algorithm cforest. In the following paper it is described (towards the end) how to use the different R packages to measure variable importance https://journal.r-project.org/archive/2009-2/RJournal_2009-2_Strobl~et~al.pdf