FMradio (version 1.1.1)

RF: Redundancy filtering of a square (correlation) matrix

Description

RF is a function that performs redundancy filtering (RF) of a square (correlation) matrix.

Usage

RF(R, t = .95)

Arguments

R

Square (correlation) matrix.

t

A scalar numeric indicating the absolute value for thresholding.

Value

Returns a redundancy-filtered matrix.

Details

Radiomic features can be very strongly correlated. The sample correlation matrix on extracted radiomic features will then often display strong collinearity. The collinearity may be so strong as to imply redundant information, in the sense that some entries will approach perfect (negative) correlation. Hence, one may wish to perform redundancy-filtering on the raw sample correlation matrix in such situations.

The RF function uses an Algorithm from Peeters et al. (2019) to remove the minimal number of redundant features under absolute marginal correlation threshold t. We recommend setting \(\mathrm{t} \in [.9,.95]\). Details of the algorithm can be found in Peeters et al. (2019).

The function returns a redundancy-filtered correlation matrix. This return output may subsequently be used in the subSet function. This is a convenience function that subsets a dataset to the features retained after redundancy-filtering.

References

Peeters, C.F.W. et al. (2019). Stable prediction with radiomics data. arXiv:1903.11696 [stat.ML].

See Also

subSet, regcor

Examples

# NOT RUN {
## Generate some (high-dimensional) data
## Get correlation matrix
p = 25
n = 10
set.seed(333)
X = matrix(rnorm(n*p), nrow = n, ncol = p)
colnames(X)[1:25] = letters[1:25]
R <- cor(X)

## Redundancy visualization, at threshold value .9
radioHeat(R, diag = FALSE, threshold = TRUE, threshvalue = .9)

## Redundancy-filtering of correlation matrix
Rfilter <- RF(R, t = .9)
dim(Rfilter)
# }