rf.ImpScale: Scaling of Random Forests importance values

In jeffreyevans/rfUtilities: Random Forests Model Selection and Performance Evaluation

Scaling of Random Forests importance values

Description

Various scaling approaches for scaling permuted parameter importance values

Usage

rf.ImpScale(x, scaling = c("mir", "se", "p"), n = 99, sort = FALSE)

Arguments

x	randomForest or ranger object with a classification or regression instance
scaling	Type of importance scaling, options are ("mir","se", "p")
n	If scaling = "p" how many permutations
sort	Sort output by importance

Details

The "mir" scale option performs a row standardization and the "se" option performs normalization using the "standard errors" of the permutation-based importance measure. Both options result in a 0-1 range but, "se" sums to 1.

The scaled importance measures are calculated as: mir = i/max(i) and se = (i / se) / ( sum(i) / se).

The "p" scale option calculates a p-value using the Janitza et al, (2018) method where, assuming that noise variables vary randomly around zero, uses negative values to build a null distribution.

Value

A data.frame with:

• "parameter" Name of the parameter

• "importance" scaled importance values

• "pvalue" Optional p-value for ranger objects

Author(s)

Jeffrey S. Evans <[e-mail]>

References

Altmann, A., Tolosi, L., Sander, O. & Lengauer, T. (2010). Permutation importance: a corrected feature importance measure, Bioinformatics 26:1340-1347.

Murphy M.A., J.S. Evans, and A.S. Storfer (2010) Quantify Bufo boreas connectivity in Yellowstone National Park with landscape genetics. Ecology 91:252-261

Evans J.S., M.A. Murphy, Z.A. Holden, S.A. Cushman (2011). Modeling species distribution and change using Random Forests CH.8 in Predictive Modeling in Landscape Ecology eds Drew, CA, Huettmann F, Wiersma Y. Springer

Janitza, S., E. Celik, and A.-L. Boulesteix (2018). A computationally fast variable importance test for random forests for high-dimensional data. Advances in Data Analysis and Classification 12(4):885-915

See Also

importance_pvalues details on Altmann's p-value method

Examples

library(randomForest)
library(ranger)

#### Classification
data(iris)
iris$Species <- as.factor(iris$Species)

# Random Forests using x, y arguments with index
( rf.class <- randomForest(x = iris[,1:4], y = iris[,"Species"], 
                           importance = TRUE) )

  rf.ImpScale(rf.class, scaling="mir")
  rf.ImpScale(rf.class, scaling="se")

# ranger using formula interface 
( rgr.class <- ranger(Species ~., data=iris, 
                      importance = "permutation") )
				   
  rf.ImpScale(rgr.class, scaling="mir")
  rf.ImpScale(rgr.class, scaling="p", n=10)

#### Regression
data(airquality)
  airquality <- na.omit(airquality)
  
( rf.reg <- randomForest(x=airquality[,2:6], y=airquality[,1],
                         importance = TRUE) )
  rf.ImpScale(rf.reg, scaling="mir")
  rf.ImpScale(rf.reg, scaling="se")

( rgr.reg <- ranger(x=airquality[,2:6], y=airquality[,1],
                    importance = "permutation") )
  rf.ImpScale(rgr.reg, scaling="mir")
  rf.ImpScale(rgr.reg, scaling="p", n=10)

jeffreyevans/rfUtilities documentation built on Nov. 12, 2023, 6:52 p.m.