Label Noise Filters for Data Preprocessing in Classification

An extensive implementation of state-of-the-art and classical algorithms to preprocess label noise in classification problems.

NoiseFiltersR contains an extensive implementation of state-of-the-art and classical label noise preprocessing algorithms for classification problems. Such a collection was missing for R statistical software.

Namely, NoiseFiltersR includes 30 label noise filters. All of them are appropriately documented, with a general explanation of the method and the exact reference where it was first published. Moreover, they can be called in a R-user-friendly manner, and their results are unified by means of the filter class, which also benefits from adapted print and summary methods.

Use install.packages to install NoiseFiltersR and its dependencies from CRAN:


Once installed, use the command library to attach the package:


Once the package is installed and attached, the user can apply any of the implemented algorithms.

Next instruction shows how to use the well-known Iterative Partitioning Filter (IPF) (Khoshgoftaar & Rebours, 2007) to filter out class noise from the dataset iris. The formula allows us to indicate the classification variable. Default parameters for the algorithm are considered:

out <- IPF(Species~., data = iris)

Then, the variable out is an object of class filter. This is a list with seven elements:

  • cleanData: a data frame containing the filtered dataset.
  • remIdx: a vector of integers indicating the indexes for removed instances (i.e. their row number with respect to the original data frame).
  • repIdx: a vector of integers indicating the indexes for repaired/relabelled instances (i.e. their row number with respect to the original data frame).
  • repLab: a factor containing the new labels for repaired instances.
  • parameters: a list containing the tuning parameters used for the filter.
  • call: an expression that contains the original call to the filter.
  • extraInf: a character that includes additional relevant information not covered by previous items.

To appropriately display the information contained in a filter object, general functions print and summary can be used (more details about their output can be found in the package vignette):


Finally, all the implemented algorithms can also be used without a formula argument, just indicating the dataset to be preprocessed and the column that contains the classification variable (last column is assumed by default):

out <- IPF(iris, classColumn = 5)

For more specific information on how to use each filter, please refer to the functions documentation page and the examples contained therein. For a general overview of the NoiseFiltersR package, please look up the associated vignette.


NoiseFiltersR 0.1.0

First release of NoiseFiltersR. This version includes:

  • An extensive implementation and documentation of state-of-the-art and classical label noise preprocessing algorithms for classification problems.

  • A specific class for filtered datasets, as well as methods for printing and summarizing the related information.

  • Functionality to indicate the classification variable by means of a formula, as is common in most of R implementations.

  • A vignette that provides the user with a general overview of the package.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


0.1.0 by Julian Luengo, 8 months ago

Browse source code at

Authors: Pablo Morales [aut], Julian Luengo [aut, cre], Luis P.F. Garcia [aut], Ana C. Lorena [aut], Andre C.P.L.F. de Carvalho [aut], Francisco Herrera [aut]

Documentation:   PDF Manual  

GPL-3 license

Imports RWeka, kknn, nnet, caret, e1071, rpart, randomForest, MASS, rJava, stats, utils

Suggests knitr, rmarkdown, testthat

See at CRAN