A divide and conquer approach for imbalanced multi-class classification and its application to medical decision...

Area of Science:

• Machine Learning
• Data Science
• Bioinformatics

Background:

• Real-world datasets often exhibit imbalanced class distributions and multiple categories.
• This imbalance is particularly prevalent in medical diagnosis and pharmaceutical testing, where rare events can have significant consequences.
• Existing research often addresses multi-class classification and imbalanced data separately, leaving a gap in combined approaches.

Purpose of the Study:

• To develop and evaluate a novel approach for imbalanced multi-class classification.
• To specifically address challenges in medical diagnosis and pharmaceutical data analysis.
• To improve classification performance on datasets with disparate category sizes.

Main Methods:

• A 'divide and conquer' strategy to partition multi-class data.
• A self-adaptive data resampling technique to manage class imbalance.
• Validation on 23 diverse UCI datasets, including medical and pharmaceutical domains.

Main Results:

• The proposed methods demonstrated superior performance compared to existing techniques.
• Significant improvements were observed, especially on medical and pharmaceutical datasets.
• The approach effectively handles datasets with a high degree of class imbalance and multiple classes.

Conclusions:

• The developed 'divide and conquer' and self-adaptive resampling methods offer an effective solution for imbalanced multi-class classification.
• This approach shows particular promise for applications in medical diagnosis and pharmaceutical research.
• The findings suggest a new direction for handling complex, imbalanced datasets in scientific research.