Calculate Weight-of-Evidence

woe() builds the Weight-of-Evidence ("WoE") dictionary of a set of predictor variables upon a given binary outcome.

Usage

woe(
  data,
  outcome,
  predictors,
  Laplace = 1e-06,
  verbose = TRUE,
  method = c("dict", "add", "replace")
)

Arguments

data: A data frame containing the outcome variable and any variables passed to predictors
outcome: The column variable in the data data frame representing the outcome (i.e., the dependent variable in data); must have exactly 2 distinct values
predictors: <tidy-select> The column variable(s) in the data data frame representing the independent variable(s) (i.e., "predictor" variable(s)); omitting this argument will select all column variables in data except outcome
Laplace: (Numeric) The pseudocount parameter of the Laplace Smoothing estimator; default is 1e-6. This value helps to avoid -Inf/Inf WoE values from arising in situations where a independent variable class has only one outcome class in the data. Set to 0 to allow Inf/-Inf WoE values.
verbose: (Logical) Should information on the WoE calculation be printed in the console?
method: (String) One of "dict", "add", or "replace". Default is "dict", which creates a Weight-of-Evidence dictionary for each unique class across each variable in predictors. "add" appends new "woe_*" columns to the input data frame (data), while "replace" replaces the variables passed to predictors with the Weight-of-Evidence equivalents.

Value

A tibble with calculated Weights-of-Evidence

Details

Negative "WoE" values indicate that there is a higher proportion of "bads" than "goods" in that particular class of the independent variable, while positive "WoE" values indicate that there is a higher proportion of "goods" than there are "bads". This hinges on the assumption that the levels of the outcome variable are ordered c("good", "bad"), otherwise the inverse is true.

References

Siddiqi, Naeem (2017). Intelligent Credit Scoring: Building and Implementing Better Credit Risk Scorecards. 2nd ed., Wiley., p. 184

Hvitfeldt E, Kuhn M (2022). embed: Extra Recipes for Encoding Predictors. https://embed.tidymodels.org, https://github.com/tidymodels/embed.

Examples

# View the order of levels in `loans$default_status`
levels(loans$default_status)
#> [1] "bad"  "good"

# Reverse levels in dependent variable for WoE calculation
df <- loans |>
  dplyr::mutate(
    default_status = factor(default_status, levels = c("good", "bad"))
  )

# Create a WoE dictionary
df |>
  woe(
    outcome = default_status,
    predictors = c(industry, housing_status)
  )
#> 
#> ── `outcome` variable `default_status` has two levels: 
#> • Using "good" as *good* level.
#> • Using "bad" as *bad* level.
#> 
#> ── You can reverse this by running: 
#> df$default_status <- factor(df$default_status, levels = c("bad", "good"))
#> 
#> ℹ See ?woe for further detail
#> # A tibble: 12 × 8
#>    variable       class        n_total n_good n_bad p_good   p_bad     woe
#>    <chr>          <chr>          <int>  <int> <int>  <dbl>   <dbl>   <dbl>
#>  1 industry       ""                 9      8     1 0.0114 0.00333  1.23  
#>  2 industry       "beef"            97     63    34 0.09   0.113   -0.231 
#>  3 industry       "dairy"          181    123    58 0.176  0.193   -0.0956
#>  4 industry       "fruit"          234    145    89 0.207  0.297   -0.359 
#>  5 industry       "grain"          280    218    62 0.311  0.207    0.410 
#>  6 industry       "greenhouse"      12      7     5 0.01   0.0167  -0.511 
#>  7 industry       "nuts"            22     14     8 0.02   0.0267  -0.288 
#>  8 industry       "pork"            50     28    22 0.04   0.0733  -0.606 
#>  9 industry       "poultry"        103     86    17 0.123  0.0567   0.774 
#> 10 industry       "sod"             12      8     4 0.0114 0.0133  -0.154 
#> 11 housing_status "own"            713    527   186 0.753  0.62     0.194 
#> 12 housing_status "rent"           287    173   114 0.247  0.38    -0.430