How we can find out that which features (predictors) play more significant roles as compared to all others used for making the final prediction in a random forest? Suppose I used 10 features for training, I want to know which 2 among them are more significant ones.
.
You can have a look at
http://arxiv-web3.library.cornell.edu/abs/1310.5726
and the references therein to the permutation importance measure
(among many others ! Just google :"variable selection" "random forest")
.
Install R, get the randomForest package, and run this code:
library(randomForest)
set.seed(4543)
# Random Forest relative importance of variables as predictors
rffit
The attached articles discuss importance measures of variables in RF. You may find them useful for your research
http://www.sciencedirect.com/science/article/pii/S0167947307003076
http://www.sciencedirect.com/science/article/pii/S0167865510000954
https://hal.inria.fr/file/index/docid/755489/filename/PRLv4.pdf
Use Variable Importance Plot in randomForest.
Description
Dotchart of variable importance as measured by a Random Forest
Usage
varImpPlot(x, sort=TRUE, n.var=min(30, nrow(x$importance)), type=NULL, class=NULL, scale=TRUE, main=deparse(substitute(x)), ...)
Arguments
x
An object of class randomForest.
sort
Should the variables be sorted in decreasing order of importance?
n.var
How many variables to show? (Ignored if sort=FALSE.)
type, class, scale
arguments to be passed on to importance
main
plot title.
...
Other graphical parameters to be passed on to dotchart.
Value
Invisibly, the importance of the variables that were plotted.
Author(s)
Andy Liaw andy\[email protected]
See Also
randomForest, importance
Examples
set.seed(4543) data(mtcars) mtcars.rf
- Badges
- Science topic