mRMR_disc {dprep}R Documentation

Feature selection for discrete attributes using the minimum redundancy and Maximum relevance method.

Description

This function selects the best predictors using the minimum redundancy and Maximum relevance method.

Usage

mRMR_disc(datos, size = 3, criterion = c("MIQ", "MID"))

Arguments

datos

The dataset

size

The number of featires to be selected

criterion

Either MIQ or MID

Details

This a hybrid feature selection method that chooses as the best features as the one that have high Mutual Infomation with the classes (V) and low mutual information among them (W). These two measures are combined two criteria: MIQ= that maximizes V/w and MID that maximizes V-W.

Value

selec_list

List of best features for prediction

Author(s)

Edgar Acuna

References

Peng, H.C., Long, F., and Ding, C. Feature selection based on mutual information. criteria of max-dependency, max-relevance, and min-redundancy.

See Also

mRMR_cont

Examples

data(bupa)
bupad=disc.ew(bupa,1:6,out="num")
mRMR_disc(bupad,3,criterion="MIQ")

[Package dprep version 3.1.1 Index]