RSMOTE: improving classification performance over imbalanced medical datasets.

Area of Science:

• Machine Learning
• Medical Informatics
• Data Science

Background:

• Medical diagnosis relies on accurate data, but imbalanced datasets can introduce bias.
• The Synthetic Minority Over-sampling Technique (SMOTE) aims to balance datasets but can lead to class mixture and reduced performance.
• Existing SMOTE modifications struggle with over-generalization.

Purpose of the Study:

• To introduce RSMOTE, a modified SMOTE method, to address shortcomings in medical imbalanced datasets.
• To analyze and verify the performance of RSMOTE on imbalanced medical data.

Main Methods:

• RSMOTE divides minority samples into four regions (normal, semi-normal, semi-critical, critical) based on density analysis.
• It globally identifies minority sample regions and applies resampling to specific sample groups.

Main Results:

• Generating synthetic samples in high-density minority regions improves classification performance by minimizing class mixture.
• RSMOTE overcomes over-generalization by globally determining minority sample regions.
• Experiments across various imbalanced medical datasets demonstrated improved effectiveness using metrics like Recall, Precision, and ROC area.

Conclusions:

• RSMOTE effectively addresses imbalanced medical datasets by segmenting minority samples into four density-based regions.
• Resampling in high-density minority regions yields superior results compared to low-density regions.
• RSMOTE offers a flexible approach to generating synthetic samples with varying minority densities for improved medical data analysis.