Apply SMOTEN Algorithm

step_smoten() creates a specification of a recipe step that generate new examples of the minority class using nearest neighbors of these cases, for data sets where all predictors are categorical (nominal). The Value Difference Metric (VDM) is used to measure the distance between observations. For each predictor, the most common category among neighbors is chosen.

Usage

step_smoten(
  recipe,
  ...,
  role = NA,
  trained = FALSE,
  column = NULL,
  over_ratio = 1,
  neighbors = 5,
  indicator_column = NULL,
  skip = TRUE,
  seed = sample.int(10^5, 1),
  id = rand_id("smoten")
)

Arguments

recipe

A recipe object. The step will be added to the sequence of operations for this recipe.

...

One or more selector functions to choose which variable is used to sample the data. See recipes::selections for more details. The selection should result in single factor variable. For the tidy method, these are not currently used.

role

Not used by this step since no new variables are created.

trained

A logical to indicate if the quantities for preprocessing have been estimated.

column

A character string of the variable name that will be populated (eventually) by the ... selectors.

over_ratio

A numeric value for the ratio of the minority-to-majority frequencies. The default value (1) means that all other levels are sampled up to have the same frequency as the most occurring level. A value of 0.5 would mean that the minority levels will have (at most) (approximately) half as many rows as the majority level.

A named numeric vector can be used instead to give different levels different targets, for example c(a = 1, b = 0.5). The names must be levels of the outcome and the values are ratios of the majority level, exactly as in the single-number case. Levels that are not named are left untouched, as are rows with a missing outcome. Because a vector of targets is not a single value, supplying one means this argument can no longer be tuned. See vignette("ratio", package = "themis") for more details.

neighbors

An integer. Number of nearest neighbor that are used to generate the new examples of the minority class.

indicator_column

A single string or NULL (the default). If a string is given, a logical column with that name is added to the output, marking rows added by the step (TRUE) vs rows from the original data (FALSE).

skip

A logical. Should the step be skipped when the recipe is baked by bake()? While all operations are baked when prep() is run, some operations may not be able to be conducted on new data (e.g. processing the outcome variable(s)). Care should be taken when using skip = TRUE as it may affect the computations for subsequent operations.

seed

An integer that will be used as the seed when applied.

id

A character string that is unique to this step to identify it.

Value

An updated version of recipe with the new step added to the sequence of existing steps (if any). For the tidy method, a tibble with columns terms which is the variable used to sample.

Details

SMOTEN is a variant of SMOTE designed for data sets where all predictors are categorical (nominal). Since standard SMOTE relies on interpolation in continuous space, it cannot be applied to categorical features directly. SMOTEN instead uses the Value Difference Metric (VDM) to measure the distance between observations. For each minority class example, new synthetic examples are generated by taking the most common value of each predictor among its nearest neighbors. The number of new examples generated is controlled by over_ratio.

All columns in the data are sampled and returned by recipes::juice() and recipes::bake().

All predictor columns must be categorical (factor or character) with no missing data.

When used in modeling, users should strongly consider using the option skip = TRUE so that the extra sampling is not conducted outside of the training set.

Minimum observations

Each minority class must have at least neighbors + 1 observations to perform the SMOTEN algorithm.

Value Difference Metric

The Value Difference Metric (VDM) used here deviates from Chawla's stated form in two ways. The per-feature deltas are aggregated by summing them (r = 1) rather than by taking their Euclidean norm (r = 2), and when a synthetic value is chosen by majority vote of the nearest neighbors the seed observation itself is excluded from the vote. The metric is internally consistent and is a valid VDM variant, but be aware of these choices when comparing results with other implementations.

Tidying

When you tidy() this step, a tibble is returned with columns terms and id:

terms: character, the selectors or variables selected
id: character, id of this step

Tuning Parameters

This step has 2 tuning parameters:

over_ratio: Over-Sampling Ratio (type: double, default: 1)
neighbors: # Nearest Neighbors (type: integer, default: 5)

Case weights

The underlying operation does not allow for case weights. Supplying data with a case weights column to this step results in an error.

References

Chawla, N. V., Bowyer, K. W., Hall, L. O., and Kegelmeyer, W. P. (2002). Smote: Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16:321-357.

Examples

library(recipes)
library(modeldata)
data(hpc_data)

hpc_cat <- hpc_data[, c("class", "protocol", "day")]

orig <- count(hpc_cat, class, name = "orig")
orig
#> # A tibble: 4 × 2
#>   class  orig
#>   <fct> <int>
#> 1 VF     2211
#> 2 F      1347
#> 3 M       514
#> 4 L       259

up_rec <- recipe(class ~ ., data = hpc_cat) |>
  step_smoten(class) |>
  prep()

training <- up_rec |>
  bake(new_data = NULL) |>
  count(class, name = "training")
training
#> # A tibble: 4 × 2
#>   class training
#>   <fct>    <int>
#> 1 VF        2211
#> 2 F         2211
#> 3 M         2211
#> 4 L         2211

# Since `skip` defaults to TRUE, baking the step has no effect
baked <- up_rec |>
  bake(new_data = hpc_cat) |>
  count(class, name = "baked")
baked
#> # A tibble: 4 × 2
#>   class baked
#>   <fct> <int>
#> 1 VF     2211
#> 2 F      1347
#> 3 M       514
#> 4 L       259