Our methods can be easily understood with the following simulated data set, Circle,
which has two classes. Fig. 1 (a) shows the original distribution of the data set,
the circle points represent majority examples and the plus signs are minority examples.
Firstly, we apply borderline-SMOTE to find out the borderline examples of the
minority class, which are denoted by solid squares in Fig. 1 (b). Then, new synthetic
examples are generated through those borderline examples of the minority class. The
synthetic examples are shown in Fig. 1 (c) with hollow squares. It is easy to find out
from the figures that, different from SMOTE, our methods only over-sample or
strengthen the borderline and its nearby points of the minority class.