imbalanceDatRel (version 0.1.5)

DatRel: Data Relocation for Resampled Data using Pure and Proper Class Cover Catch Digraph

Description

DatRel relocates resampled data using Pure and Proper Class Cover Catch Digraph

Usage

DatRel(x, y, x_syn, proportion = 1, p_of = 0, class_pos = NULL)

Value

an list object which includes:

x_new

Oversampled and relocated feature matrix

y_new

Oversampled class variable

x_syn

Generated and relocated sample matrix

i_dominant

Indexes of dominant samples

x_pos_dominant

Dominant samples for positive class

radii_pos_dominant

Positive class cover percentage

Arguments

x

feature matrix or dataframe.

y

class factor variable.

x_syn

synthetic data generated by an oversampling method.

proportion

proportion of covered samples. A real number between \((0,1]\). 1 by default. Algorithm stops when desired percent of coverage achieved in each class. Smaller numbers results in less dominant samples.

p_of

proportion to increase cover radius. A real number between \((0,\infty)\). Default is 0. Higher values tolerate other classes more.

class_pos

Class name of synthetic data. Default is NULL. If NULL, positive class is minority class.

Author

Fatih Saglam, [e-mail]

Details

Calculates cover areas using pure and proper class cover catch digraphs (PCCCD) for original dataset. Any sample outside of cover area is relocated towards a specific dominant point. Determination of dominant point to move towards is based on distance based on radii of PCCCD balls. p_of is to increase obtained radii to be more tolerant to noise. prooportion argument is cover percentage for PCCCD to stop when desired percentage is covered for each class. PCCCD models are determined using rcccd package. class_pos argument is used to specify oversampled class.

Examples

library(SMOTEWB)
library(rcccd)

set.seed(10)
# adding data
x <- rbind(matrix(rnorm(2000, 3, 1), ncol = 2, nrow = 1000),
           matrix(rnorm(60, 6, 1), ncol = 2, nrow = 30))
y <- as.factor(c(rep("negative", 1000), rep("positive", 30)))

# adding noise
x[1001,] <- c(3,3)
x[1002,] <- c(2,2)
x[1003,] <- c(4,4)

# resampling
m_SMOTE <- SMOTE(x = x, y = y, k = 3)

# relocation of resampled data
m_DatRel <- DatRel(x = x, y = y, x_syn = m_SMOTE$x_syn)

# resampled data
plot(x, col = y, main = "SMOTE")
points(m_SMOTE$x_syn, col = "green")

# resampled data after relocation
plot(x, col = y, main = "SMOTE + DatRel")
points(m_DatRel$x_syn, col = "green")