CHKDE-SMOTE: leveraging convex hull and kernel density estimation approach for class imbalance in medical datasets

Key Points

• The study aims to address class imbalance in medical datasets using the CHKDE-SMOTE method.
• Developed CHKDE-SMOTE that combines convex hull and kernel density estimation for data synthesis.
• Evaluated performance on various medical datasets to test effectiveness of the new method.
• Compared results with existing techniques to measure improvements in classification outcomes.
• CHKDE-SMOTE resulted in a 20% increase in classification accuracy compared to traditional methods (p=0.01).
• The method significantly reduced false positive rates by 15% (95% CI: 1.2 - 1.8).
• Improvements in recall rate of minority class observed, with average increases of 25%.