CORE: core-based synthetic minority over-sampling and borderline majority under-sampling technique.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Class imbalance is a significant challenge in machine learning.
• Traditional algorithms struggle with imbalanced datasets, failing to detect minority classes.
• The rare class is often the primary interest in classification tasks.

Purpose of the Study:

• To propose a novel technique, CORE (Core-Oriented Risk Elimination), to address class imbalance.
• To enhance the discriminative power of the minority class in imbalanced datasets.
• To improve the predictive performance for minority classes.

Main Methods:

• CORE strengthens the core of the minority class.
• It weakens the risk of misclassifying minority instances near the majority class borderline.
• Safe levels are used to define core and borderline regions.

Main Results:

• The CORE technique makes the minority class more clustered and dominant.
• Experiments demonstrate significant improvements in minority class predictive performance.
• CORE effectively handles datasets with severe class imbalance.

Conclusions:

• CORE is an effective method for tackling class imbalance problems.
• The technique enhances the ability to detect and classify minority classes.
• CORE offers a promising solution for real-world imbalanced classification scenarios.