Skip to contents

Find Highly Correlated Columns

Source: R/corr_test_all.R
drop_correlated_columns.Rd

Identifies numeric columns to drop for multicollinearity, via findCorrelation on the pairwise correlation matrix of data's numeric columns. Named drop_correlated_columns() (not findCorrelation()) so it doesn't shadow caret's function of the same name for anyone with both packages loaded.

Usage

drop_correlated_columns(
  data,
  use = "pairwise.complete.obs",
  method = "pearson",
  cutoff = 0.9,
  verbose = FALSE,
  names = TRUE,
  exact = ncol(data) < 100
)

Arguments

data

A data frame or tibble.

use

Passed to cor (default "pairwise.complete.obs").

method

Passed to cor (default "pearson").

cutoff

Absolute correlation above which a column is flagged (default 0.9).

verbose

Logical; passed to findCorrelation.

names

Logical; if TRUE (default) return column names instead of indices.

exact

Passed to findCorrelation; defaults to ncol(data) < 100 (the original had a bug here referencing an undefined x instead of data — fixed).

Value

Character vector (or integer indices) of columns to remove.

Details

**Note:** `drop_correlated_columns()` identifies column names or indices to *remove* to mitigate multicollinearity. To inspect the pairwise correlation coefficients themselves in a tidy table, use [correlated_pairs_table()] or [corr_test_all()].

See also

[correlated_pairs_table()], [corr_test_all()], [plot_corr_scatter()]

Examples

if (requireNamespace("caret", quietly = TRUE)) {
  drop_correlated_columns(mtcars, cutoff = 0.8)
}
#> [1] "cyl"  "disp" "mpg"