SVM-SMOTE generates new examples of the minority class near the decision boundary, using the support vectors of a fitted SVM to decide where to place synthetic examples.
svmsmote(
df,
var,
k = 5,
over_ratio = 1,
distance = "euclidean",
m_neighbors = NULL,
out_step = 0.5
)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.
An integer. Number of nearest neighbor that are used to generate the new examples of the minority class.
A numeric value for the ratio of the minority-to-majority frequencies. The default value (1) means that all other levels are sampled up to have the same frequency as the most occurring level. A value of 0.5 would mean that the minority levels will have (at most) (approximately) half as many rows as the majority level.
A named numeric vector can be used instead to give different levels
different targets, for example c(a = 1, b = 0.5). The names must be
levels of the outcome and the values are ratios of the majority level,
exactly as in the single-number case. Levels that are not named are left
untouched, as are rows with a missing outcome. Because a vector of targets
is not a single value, supplying one means this argument can no longer be
tuned. See vignette("ratio", package = "themis") for more details.
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.
An integer or NULL. Number of nearest neighbors, among
all classes, that are used to label each minority support vector as noise,
danger, or safe. Defaults to NULL, which means 2 * k.
A number. Step size used when extrapolating new examples away from safe support vectors. Defaults to 0.5.
SVM-SMOTE (Support Vector Machine SMOTE) works the same way as SMOTE, except that instead of generating points around every point of the minority class, it focuses generation near the decision boundary. A support vector machine is fitted to the data and the support vectors that belong to the minority class are used as the base points for generating new examples.
For each minority support vector its nearest neighbors among all classes are calculated. If all of the neighbors come from a different class the support vector is labeled noise and is discarded. If more than half of the neighbors come from a different class the support vector is labeled "danger" and new points are interpolated between it and its minority-class neighbors. The remaining support vectors are considered to be in a safe region and new points are extrapolated away from their minority-class neighbors.
The number of neighbors used for this labeling is controlled by
m_neighbors, and how far the extrapolated points are placed is controlled by
out_step.
The support vector machine is always fitted with kernlab::ksvm() using a
radial basis function kernel (kernel = "rbfdot") and a cost of C = 1,
matching the reference implementations of the method. These are not exposed
as arguments because the fitted model is only used to identify which minority
observations are support vectors, and not for prediction, so the sampling
results are relatively insensitive to them.
SMOTE generates new examples of the minority class using nearest neighbors
of these cases. For each existing minority class example, new examples are
created by interpolating between the example and its nearest neighbors. The
number of nearest neighbors used is controlled by the number of neighbors
argument (k in smote(), neighbors in step_smote()), and the number
of new examples generated is controlled by over_ratio.
All columns used in this function must be numeric with no missing data.
Nguyen, H. M., Cooper, E. W., and Kamei, K. (2011). Borderline over-sampling for imbalanced data classification. International Journal of Knowledge Engineering and Soft Data Paradigms, 3(1), 4-21.
step_svmsmote() for step function of this method
Other Direct Implementations:
adasyn(),
bsmote(),
cluster_centroids(),
cnn(),
enn(),
instance_hardness(),
kmeans_smote(),
ncl(),
nearmiss(),
oss(),
rose(),
smogn(),
smote(),
smoten(),
smotenc(),
tomek()
if (FALSE) { # rlang::is_installed("kernlab")
circle_numeric <- circle_example[, c("x", "y", "class")]
res <- svmsmote(circle_numeric, var = "class")
res <- svmsmote(circle_numeric, var = "class", k = 10)
res <- svmsmote(circle_numeric, var = "class", over_ratio = 0.8)
res <- svmsmote(circle_numeric, var = "class", distance = "manhattan")
res <- svmsmote(circle_numeric, var = "class", m_neighbors = 20)
}
Run the code above in your browser using DataLab