born

The born package provides new methods for the artificial addition of label noise in the borderline examples to add more challenging and yet realistic artificial noise to classification datasets. The main difference between the methods is the criterion and bias adopted to estimate which are the borderline examples to be disturbed: the first method is based on the ratio of intra/inter class Nearest Neighbor distance and the second method is based on the distance between the examples and the decision border induced by a radial kernel.

Installation

The installation process using devtools is:

if (!require("devtools")) {
    install.packages("devtools")
}
devtools::install_github("lpfgarcia/born")
library("born")

Example of use

The simplest way to generate the noisy dataset is using the random, neighborwise and nonlinearwise functions. The methods can be called by a symbolic description of the dataset (formula) or by a data frame (x, y). The parameters are the dataset and the ratio of noise. A simple example is given next:

## Generate an iris dataset with 10% of random noise using formula
random(Species~., iris, rate=0.1)

## Generate an iris dataset with 10% of random noise using x and y
random(x=iris[,1:4], y=iris[,5], rate=0.1)

## Generate an iris dataset with 10% of neighborwise noise
neighborwise(Species~., iris, rate=0.1)

## Generate an iris dataset with 10% of nonlinearwise noise
nonlinearwise(Species~., iris, rate=0.1)

Developer notes

To cite born in publications use: Luis Garcia, Jens Lehmann, Andre de Carvalho, and Ana Lorena. (2018). born: Generate Borderline Noise for Classification Problems. R package version 0.1.0, https://github.com/lpfgarcia/born/

To submit bugs and feature requests, report at project issues.