scutr (version 0.2.0)

SCUT: SMOTE and cluster-based undersampling technique.

Description

This function balances multiclass training datasets. In a dataframe with n classes and m rows, the resulting dataframe will have m / n rows per class. SCUT_parallel() distributes each over/undersampling task across multiple cores. Speedup usually occurs only if there are many classes using one of the slower resampling techniques (e.g. undersample_mclust()). Note that SCUT_parallel() will always run on one core on Windows.

Usage

SCUT(
  data,
  cls_col,
  oversample = oversample_smote,
  undersample = undersample_mclust,
  osamp_opts = list(),
  usamp_opts = list()
)
SCUT_parallel(
  data,
  cls_col,
  ncores = detectCores()%/%2,
  oversample = oversample_smote,
  undersample = undersample_mclust,
  osamp_opts = list(),
  usamp_opts = list()
)

Value

A dataframe with equal class distribution.

Arguments

data

Numeric data frame.

cls_col

The column in data with class membership.

oversample

Oversampling method. Must be a function with the signature foo(data, cls, cls_col, m, ...) that returns a data frame, one of the oversample_* functions, or resample_random().

undersample

Undersampling method. Must be a function with the signature foo(data, cls, cls_col, m, ...) that returns a data frame, one of the undersample_* functions, or resample_random().

osamp_opts

List of options passed to the oversampling function.

usamp_opts

List of options passed to the undersampling function.

ncores

Number of cores to use with SCUT_parallel().

Details

Custom functions can be used to perform under/oversampling (see the required signature below). Parameters represented by ... should be passsed via osamp_opts or usamp_opts as a list.

References

Agrawal A, Viktor HL, Paquet E (2015). 'SCUT: Multi-class imbalanced data classification using SMOTE and cluster-based undersampling.' In 2015 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K), volume 01, 226-234.

Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP (2002). 'SMOTE: Synthetic Minority Over-sampling Technique.' Journal of Artificial Intelligence Research, 16, 321-357. ISSN 1076-9757, tools:::Rd_expr_doi("10.1613/jair.953"), https://www.jair.org/index.php/jair/article/view/10302.

Examples

ret <- SCUT(iris, "Species", undersample = undersample_hclust,
            usamp_opts = list(dist_calc="manhattan"))
ret2 <- SCUT(chickwts, "feed", undersample = undersample_kmeans)
table(ret$Species)
table(ret2$feed)
# SCUT_parallel fires a warning if ncores > 1 on Windows and will run on
# one core only.
ret <- SCUT_parallel(wine, "type", ncores = 1, undersample = undersample_kmeans)
table(ret$type)