combineFS: combineFS

In enriquea/feseR: Feature Selection workflow in R

combineFS

Description

The main function controlling the Feature Selection workflow. This function combines sequencially different FS step keeping the following structure: Univariate filter -> Multivariate filter -> Wrapper method. It includes an internal cross-validation step used in the last FS step. In addition, it is posible to set up an external loop which operates over randomized and class-balanced test data. The process returns information from both training and testing phases.

Usage

combineFS(
  features,
  class,
  univariate = "corr",
  mincorr = 0.3,
  n.percent = 0.75,
  zero.gain.out = TRUE,
  multivariate = "mcorr",
  maxcorr = 0.75,
  cum.var.cutoff = 1,
  wrapper = "rfe.rf",
  number.cv = 10,
  group.sizes = c(1:10, seq(15, 100, 5)),
  extfolds = 10,
  partition = 2/3,
  metric = "Accuracy",
  tolerance = 0,
  verbose = TRUE
)

Arguments

features	A numeric matrix as input.
class	Response variable as numeric vector. It will be coerced to factor.
univariate	Descrition of the Univariate filter to be used. Set 'corr' (default) for correlation filter, 'gain' for gain information or none.
mincorr	The threshold controling the Univariate correlation filter.
n.percent	If 'gain' is selected, this parameter controls the percent of features (with higher) to be returned.
zero.gain.out	Is TRUE (default), zero-gain features will be filtered out (n.percent will be ignored).
multivariate	Multivariate filter to be used. Set 'mcorr' (default) for correlation filter, 'pca' for Principal Component Analysis or none.
maxcorr	The threshold controling the matrix correlation filter (default value 0.75).
cum.var.cutoff	If 'pca' is selected, this parameter controls the PCA process. See function filter.pca.
wrapper	Wrapper method to be used. Set 'rfe.rf' (default) for recursive feature elimination wrapped with random forest.
number.cv	See rfeRF for description.
group.sizes	See rfeRF for description.
extfolds	Number of times (default 10) to repeat the entire FS process randomizing the dataset (test/training).
partition	Parameter controling the data partition in test and training dataset. It generates random and class-balanced dataset.
metric	Metric to evaluate performance ('Accuracy' (default), 'Kappa' or 'ROC').
tolerance	Allow tolerance for evaluation metric (Default zero).
verbose	Make the output verbose.

Value

A list with the following elements.

opt.variables vector with optimal (final) features names. training dataframe with the metrics from the training phase. testing dataframe with the metrics from the testing phase. best.model the best model obtained (max. accuracy and min. number of final features). runtime the workflow runtime (secs).

enriquea/feseR documentation built on Feb. 25, 2025, 12:20 a.m.