FSmethod: Feature Selection via Matrix Correlation and Recursive...

In mrbakhsh/HPiP: Host-Pathogen Interaction Prediction

Feature Selection via Matrix Correlation and Recursive Feature Elimination (RFE)

Description

This function performs feature selections via two approaches

• filter.corr - compute matrix correlation between features and filter using a threshold.

• rfeFS - perform recursive feature elimination (RFE) method wrapped with a Random Forest (RF) algorithm for feature importance evaluation.

Usage

FSmethod(
  x,
  type = c("cor", "rfe", "both"),
  cor.cutoff = 0.7,
  resampling.method = "cv",
  iter = 2,
  repeats = 3,
  metric = "Accuracy",
  verbose = TRUE
)

Arguments

x	A data.frame containing protein-protein interactions, class labels and features.
type	The feature selection type, one or two of filter.corr and rfeFS.
cor.cutoff	Correlation coefficient cutoff used for filtering. See filter.corr for more details.
resampling.method	The resampling method for RFE :'boot', 'boot632', optimism_boot',boot_all', 'cv', 'repeatedcv', 'LOOCV', 'LGOCV';defaults to cv. See rfeFS and rfeControl for more details.
iter	Number of partitions for cross-validation; defaults to 2. See rfeFS and rfeControl for more details.
repeats	For repeated k-fold cross validation only; defaults to 3.See rfeFS and rfeControl for more details.
metric	A string that specifies what summary metric will be used to select the optimal feature ; default to ROC.See rfeFS and rfe for more details.
verbose	Make the output verbose.See rfeFS and rfeControl for more details.

Details

FSmethod

Value

If the type set to filter.corr , the output includes the following elements:

• corProfile - A correlation matrix.

• corSelectedFeatures - Name of features that retained after the correlation analysis.

• cordf - A data.frame filtered.

If the type set to rfeFS , the output includes the following elements:

• rfProfile - A list of elements. See rfe for more details.

• rfSelectedFeatures - Name of features that retained in the feature selection process.

• rfdf - A data.frame filtered.

If type set to both the output includes the following elements:

• rfdf - The final data.frame that includes the selected features retained after both filter.corr and rfeFS analysis.

Author(s)

Matineh Rahmatbakhsh, [e-mail].

Examples

data('example_data')
x <- na.omit(example_data)
s <- FSmethod(x, type = 'both',
cor.cutoff = 0.7, resampling.method = "repeatedcv",
iter = 5, repeats = 3, metric = "ROC", verbose = TRUE)

mrbakhsh/HPiP documentation built on March 28, 2023, 4:35 p.m.