[R-sig-Geo] Randmforest and VARIABLE SELECTION with leaps()
giuseppe.amatulli at gmail.com
Fri Dec 3 16:04:03 CET 2010
no, in principal you do not need to do do any variable selection or
The multiple random permutations allow to discover all the possible
relationship even if the variable are very similar. Indeed if the
variable are very similar they should be close in the variable
Usually the tree growing is not very memory demanding (for a matrix
of 100 000 ) so you do not need to reduce the number of variables. But
in case of prediction to raster data can be important to split in
tails the predictors. In this case a re-run of random forest without
the less important variables can speed up the final map creation.
On 3 December 2010 15:10, gianni lavaredo <gianni.lavaredo at gmail.com> wrote:
> Dear Researchers,
> i am not a RandomForest expertise and sorry for not smart question.
> I have several predict variables (some conceptually really similar) and I
> wish to use randomforest in R. Is It useful to use a variable secletion
> before using RandomForest to select reduce the number of variable?
> [[alternative HTML version deleted]]
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
More information about the R-sig-Geo