Under-samples the majority classes by combining Condensed Nearest Neighbors and Tomek's links, first reducing redundant majority class observations and then removing majority class observations that form Tomek links with minority class observations.
oss(df, var, distance = "euclidean")A data.frame or tibble, depending on type of df.
data.frame or tibble. Must have 1 factor variable and remaining numeric variables.
Character, name of variable containing factor variable.
A character string specifying the distance metric used for
nearest neighbor calculations, defaulting to "euclidean". The available
metrics fall into three groups.
"euclidean", "cosine", and "mahalanobis" use approximate nearest
neighbors via the RANN package and scale well to large datasets.
"squared_chord", "matusita", "hellinger", and "bhattacharyya" are
probability-divergence measures that treat each row as a distribution over
the predictors, so they require non-negative values. "hellinger" and
"bhattacharyya" further require each row to sum to 1. All four also use
the RANN package and scale well to large datasets.
"manhattan", "chebyshev", "canberra", "soergel", "lorentzian",
"jeffreys", "topsoe", "jensen-shannon", "jensen_difference",
"taneja", and "kumar-johnson" compute an exact all-pairs distance
matrix. This takes time and memory proportional to the square of the number
of observations in a class, so these are best suited to smaller datasets.
Everything from "canberra" onwards is a probability divergence requiring
non-negative values, is provided by the philentropy package (which must be
installed separately), and in the case of "jeffreys", "taneja", and
"kumar-johnson" requires strictly positive values, since those divide by
individual predictor values.
The probability divergences are meaningful for compositional predictors such as proportions or counts normalized per observation, and are generally not appropriate for standardized predictors.
One-Sided Selection (OSS) is an under-sampling method that combines two cleaning techniques. It first applies Condensed Nearest Neighbors (CNN) to reduce the majority classes to a consistent subset that correctly classifies the data using a 1-nearest-neighbor rule, discarding redundant interior observations. It then applies Tomek's links to the remaining observations, removing the majority class observations that form Tomek links with minority class observations, cleaning the decision boundary.
The smallest class is treated as the minority class and is always kept. Because the CNN step relies on a random seed observation and a random scan order, results depend on the random seed.
With more than two classes, the Tomek's links step removes both members of a majority-majority link, not only links between a majority and the minority class. The binary case, the primary intended use, is unaffected.
All columns used in this function must be numeric with no missing data.
Kubat, M., & Matwin, S. (1997). Addressing the curse of imbalanced training sets: one-sided selection. In ICML (Vol. 97, pp. 179-186).
step_oss() for step function of this method
Other Direct Implementations:
adasyn(),
bsmote(),
cluster_centroids(),
cnn(),
enn(),
instance_hardness(),
kmeans_smote(),
ncl(),
nearmiss(),
rose(),
smogn(),
smote(),
smoten(),
smotenc(),
svmsmote(),
tomek()
circle_numeric <- circle_example[, c("x", "y", "class")]
res <- oss(circle_numeric, var = "class")
res <- oss(circle_numeric, var = "class", distance = "manhattan")
Run the code above in your browser using DataLab