# Filter a set of correlated predictors to fewer ones

Source:`R/utils-predictors.R`

`predictor_filter.Rd`

This function helps to remove highly correlated variables from a set of predictors. It supports multiple options some of which require both environmental predictors and observations, others only predictors.

Some of the options require different packages to be pre-installed, such as
`ranger`

or `Boruta`

.

## Arguments

- env
A

`data.frame`

or`matrix`

with extracted environmental covariates for a given species.- keep
A

`vector`

with variables to keep regardless. These are usually variables for which prior information is known.- method
Which method to use for constructing the correlation matrix (Options:

`'pearson'`

(Default),`'spearman'`

|`'kendal'`

),`"abess"`

, or`"boruta"`

.- ...
Other options for a specific method

## Value

`vector`

of variable names to be excluded. If the
function fails due to some reason return `NULL`

.

## Details

Available options are:

`"none"`

No prior variable removal is performed (Default).`"pearson"`

,`"spearman"`

or`"kendall"`

Makes use of pairwise comparisons to identify and remove highly collinear predictors (Pearson's`r >= 0.7`

).`"abess"`

A-priori adaptive best subset selection of covariates via the`abess`

package (see References). Note that this effectively fits a separate generalized linear model to reduce the number of covariates.`"boruta"`

Uses the`Boruta`

package to identify non-informative features.