Figure 2.

(a) An example of 2D binary class data with data imbalance: majority class of black circles and minority class of red crosses. (b) Among the minority class instances, a ‘DANGER’ set was made by selecting those instances whose neighbors are more included in the majority class than in the minority class—borderline minority instances, marked by blue squares. The number of neighbors, K = 5. (c) K new synthetic instances are created for each instance in the DANGER set (see the text for more details of the synthesis procedure); newly synthesized instances in the minority class are marked by green squares.