ImpSampRegress: WEighted Relevance-based Combination Strategy (WERCS)...

In UBL: An Implementation of Re-Sampling Approaches to Utility-Based Learning for Both Classification and Regression Tasks

WEighted Relevance-based Combination Strategy (WERCS) algorithm for imbalanced regression problems

Description

This function handles imbalanced regression problems using the relevance function provided to re-sample the data set. The relevance function is used to introduce replicas of the most important examples and to remove the least important examples.

Usage

WERCSRegress(form, dat, rel = "auto", thr.rel = NA, 
               C.perc = "balance", O = 0.5, U = 0.5)

Arguments

form	A formula describing the prediction problem
dat	A data frame containing the original (unbalanced) data set
rel	The relevance function which can be automatically ("auto") determined (the default) or may be provided by the user through a matrix with interpolating points.
thr.rel	The default is NA which means that no threshold is used when performing the over/under-sampling. In this case, the over-sampling is performed by assigning a higher probability for selecting an example to the examples with higher relevance. On the other hand, the under-sampling is performed by removing the examples with less relevance. The user may chose a number between 0 and 1 indicating the relevance threshold above which a case is considered as belonging to the rare "class".
C.perc	A list containing the percentage(s) of under- or/and over-sampling to apply to each "class" obtained with the threshold. This parameter is only used when a relevance threshold (thr.rel) is set. Otherwise it is ignored. The C.perc values should be provided in ascending order of target variable values. The over-sampling percentage(s) must be numbers higher than 1 and represent the increase applied to the examples of the bump. The under-sampling percentage(s) should be numbers below 1 and represent the decrease applyed in the corresponding bump. If the value of 1 is provided for a given bump, then the examples in that bump will remain unchanged. Alternatively, this parameter may be set to "balance" (the default) or "extreme", cases where the sampling percentages are automatically estimated.
O	A number expressing the importance given to over-sampling when the thr.rel parameter is NA. When O increases the number of examples to include during the over-sampling step also increases. Default to 0.5.
U	A number expressing the importance given to under-sampling when the thr.rel parameter is NA. When U increases, the number of examples selected during the under-sampling step also increases. Defaults to 0.5.

Value

The function returns a data frame with the new data set resulting from the application of the importance sampling strategy.

Author(s)

Paula Branco [e-mail], Rita Ribeiro [e-mail] and Luis Torgo [e-mail]

See Also

RandUnderRegress, RandOverRegress

Examples

  if (requireNamespace("DMwR2", quietly = TRUE)) {
  data(algae, package ="DMwR2")
  clean.algae <- data.frame(algae[complete.cases(algae), ])
  # defining a threshold on the relevance
  IS.ext <-WERCSRegress(a7~., clean.algae, rel = "auto", 
                          thr.rel = 0.7, C.perc = "extreme")
  IS.bal <-WERCSRegress(a7~., clean.algae, rel = "auto", thr.rel = 0.7,
                          C.perc = "balance")
  myIS <-WERCSRegress(a7~., clean.algae, rel = "auto", thr.rel = 0.7,
                        C.perc = list(0.2, 6))
  # neither threshold nor C.perc defined
  IS.auto <- WERCSRegress(a7~., clean.algae, rel = "auto")
  }

UBL documentation built on Oct. 8, 2023, 1:07 a.m.