Sampling for Class Imbalances

Implements up- or down-sampling for analyses with binary class imbalances.

Usage

rebalance(data, var, method = c("down", "up"))

Arguments

data

A data frame or tibble. For example, an individual "analysis" fold (rsample::analysis()) from an rsample::vfold_cv() call.

var

character(1). A column name of data containing class labels for up-/down-sampling. Assumes a binary variable. tidyselect variable selection is also supported (see examples).

method

character(1). One of "down" or "up" (matched). Down-sampling decreases size of the major class, whereas up-sampling increases (i.e. sampling with replacement) the size per group of the minor class.

Value

A sub- or re-sampled data frame that is either smaller or larger than data with equal class balance.

Author

Stu Field

Examples

xtab <- helpr::cross_tab

# original class distribution
xtab(mtcars, vs)
#> vs
#>   0   1 Sum 
#>  18  14  32 

# down-sampling `vs`
si <- rebalance(mtcars, "vs")    # 'down'
xtab(si, vs)
#> vs
#>   0   1 Sum 
#>  14  14  28 

# up-sampling `vs` & passed variable
var <- "vs"
si <- rebalance(mtcars, var, "up")
xtab(si, vs)
#> vs
#>   0   1 Sum 
#>  18  18  36 

# also supports unquoted strings
si  <- rebalance(mtcars, vs)
xtab(si, vs)
#> vs
#>   0   1 Sum 
#>  14  14  28