featselectRF: Feature selection algorithm for Random Forests (RF)

In jjvalletta/featselectRF: Feature selection using Random Forests

Description

Implements a modified mProbes/xRF feature selection algorithm within a cross-validation loop

Usage

featselectRF(x, y, nRepeat = 100, kFold = 5, rKeep = 0.3,
  bParallel = TRUE, nThread = parallel::detectCores() - 1, nSeed = 1983,
  pCutOff = 0.05, ...)

Arguments

x	N x D predictors data frame where N - no. of samples, D - no. of features
y	a vector of factors of length N, the target class (e.g as.factor("A", "A", "B", etc.))
nRepeat	no. of times features are permuted (this is the sample size used when comparing importance score for permuted vs real features)
kFold	no. of cross-validation folds (default: 5)
rKeep	percentage of predictors to ignore after first RF fit (0>rKeep<=1) (default: 0.3)
bParallel	whether to use mProbes() or mProbesParallel() (default: True)
nThread	no. of threads to use when running in parallel (default: parallel::detected cores - 1)
nSeed	seed for cross-validation folds (default: 1983)
pCutOff	Bonferonni corrected adjusted p-value cutoff when comparing importance scores of permuted vs real predictors (default: 0.05)
...	arguments passed to the Random Forest classifier (e.g ntree, sampsize, etc.)

Value

A list with the following components:

rKeepPredictors	rKeep% predictors kept after first RF fit
topPredictors	top predictors (Bonferroni corrected p-values<pCutOff) of each fold
pValues	Bonferroni corrected pValues for rKeepPredictors
ROC	receiver operating characteristic curve, ROCR object
auc	area under the ROC curve
confMatrix	confusion matrix on test data
iiFolds	indices of the cross-validation folds

References

Huynh-Thu VA et al. Bioinformatics 2012

Nguyen et al. The Scientific World Journal 2015

Examples

bWant <- iris$Species %in% c("versicolor", "virginica")
x <- iris[bWant, 1:4]
y <- droplevels(as.factor(iris$Species[bWant]))

out <- featselectRF(x, y, nodesize=3, ntree=1001)

jjvalletta/featselectRF documentation built on May 7, 2019, 2:55 a.m.