Replies: 2 comments 3 replies
-
Generally speaking this means the model would be better off with that variable, since the prediction error is improved without it. See this vignette: https://www.randomforestsrc.org/articles/vimp.html |
Beta Was this translation helpful? Give feedback.
2 replies
-
Negative VIMP identifies noisy variables, large positive VIMP identifies strong signal variables. https://www.randomforestsrc.org/articles/rfsrc-subsample.html |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I am working on survival random forest on a dataset using package
randomForestSRC. We have 71 predictors. We got the variable importance
list and found that there is negative importance, see below for some of
them. Could anyone please help us interpret the negative importance? I
googled and couldn't find the answer. Many thanks.
CVDHISTORY | -0.387097496
RIGHTFUGFUSIFORMGYRUS_VR | -0.405789685
CEREBELLUM_VR | -0.406030867
NUMBEROFDRUGS | -0.554088997
Beta Was this translation helpful? Give feedback.
All reactions