Two density-based sampling approaches for imbalanced and overlapping data

Highlights：

• Introducing two density-based methods to achieve balance, eliminate overlap and noise data.

• Increasing minority class learning quality without significantly harming majority class learning.

• Creating a boundary between the classes while trying to maintain their structures and shapes maximally.

• Introducing a density-based hybrid sampling method to achieve balance and create a uniform distribution of data in classes.

• Comprehensive evaluation of the proposed methods in compared to other recent related works on a variant set of imbalanced datasets.