At the data level, different forms of re-sampling methods were proposed [1]. The
simplest re-sampling methods are random over-sampling and random under-sampling.
The former augments the minority class by exactly duplicating the examples of the
minority class, while the latter randomly takes away some examples of the majority
class. However, random over-sampling may make the decision regions of the learner
smaller and more specific, thus cause the learner to over-fit. Random under-sampling
can reduce some useful information of the data sets. Many improved re-sampling
methods are thus presented, such as heuristic re-sampling methods, combination of
over-sampling and under-sampling methods, embedding re-sampling methods into
data mini