Distribution-Sensitive Unbalanced Data Oversampling Method for Medical Diagnosis.

Area of Science:

• Medical Diagnostics
• Machine Learning
• Data Science

Background:

• Imbalanced datasets in medical diagnostics lead to low accuracy in classification learning algorithms.
• Existing oversampling algorithms struggle to effectively handle the nuances of imbalanced medical data.

Purpose of the Study:

• To propose a distribution-sensitive oversampling algorithm for imbalanced data in medical diagnostic applications.
• To improve the accuracy of classification learning algorithms when dealing with severely imbalanced sample sets.

Main Methods:

• The proposed algorithm categorizes minority samples into noise, unstable, boundary, and stable groups based on their location.
• It employs a distribution-sensitive sample synthesis approach, selecting methods based on sample proximity to surrounding minority data.
• This ensures synthesized samples retain characteristics of original minority samples.

Main Results:

• The algorithm demonstrates improved accuracy rates for classification learning algorithms compared to existing methods.
• Significant enhancements were observed in both accuracy and recall rates for minority classes.
• Real-world medical diagnostic data validated the effectiveness of the proposed oversampling technique.

Conclusions:

• The distribution-sensitive oversampling algorithm effectively addresses the challenge of imbalanced data in medical diagnostics.
• It offers a superior approach to enhancing classification performance, especially for underrepresented sample groups.
• The method provides a valuable tool for improving diagnostic accuracy in data-scarce medical scenarios.