rf.classBalance: Random Forest Class Balance (Zero Inflation Correction) Model

In rfUtilities: Random Forests Model Selection and Performance Evaluation

Description

Implements Evans & Cushman (2008) Random Forests class-balance (zero inflation) modeling approach.

Usage

rf.classBalance(ydata, xdata, p = 0.005, cbf = 3, sf = 2,
  seed = NULL, ...)

Arguments

ydata	Response variable using index (i.e., [,2] or [,"SPP"] )
xdata	Independent variables using index (i.e., [,3:14] or [3:ncol(data)] )
p	p-value of covariance convergence (do not recommend changing)
cbf	Scaling factor to test if problem is imbalanced, default is size of majority class * 3
sf	Majority subsampling factor. If sf=1 then random sample would be perfectly balanced with smallest class [s|0=n|1] whereas; sf=2 provides [s|0=(n|1*2)]
seed	Sets random seed in R global environment
...	Additional arguments passed to randomForest

Value

A rf.balanced object with the following components: @return model Final Combined Random Forests ensemble (randomForest object) @return OOB.error Out-of-bag error for each model (vector) @return confusion Confusion matrix for each model (list)

Note

This approach runs independent Random Forest models using random subsets of the majority class until covariance convergences on full data. The final model is obtained by combining independent ensembles.

Author(s)

Jeffrey S. Evans <jeffrey_evans<at>tnc.org>

References

Evans, J.S. and S.A. Cushman (2009) Gradient Modeling of Conifer Species Using Random Forest. Landscape Ecology 5:673-683.

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

See Also

randomForest for randomForest ... model options

Examples

require(randomForest)
data(iris)
  iris$Species <- as.character(iris$Species)
    iris$Species <- ifelse(iris$Species == "setosa", "virginica", iris$Species)
      iris$Species <- as.factor(iris$Species)	
	
# Percent of "virginica" observations
length( iris$Species[iris$Species == "virginica"] ) / dim(iris)[1]*100

# Balanced model	
( cb <- rf.classBalance( ydata=iris[,"Species"], xdata=iris[,1:4], cbf=1 ) )

# Calculate Kappa for each balanced model in ensemble 
for(i in 1:length(cb$confusion) ) { 
  print( accuracy(cb$confusion[[i]][,1:2])[5] ) 
}

# Evaluate cumulative and mean confusion matrix
accuracy( round((cb$confusion[[1]] + cb$confusion[[2]] + cb$confusion[[3]]))[,1:2] )
accuracy( round((cb$confusion[[1]] + cb$confusion[[2]] + cb$confusion[[3]])/3)[,1:2])

rfUtilities documentation built on Oct. 3, 2019, 9:04 a.m.