Ensemble of Rotation Trees for Imbalanced Medical Datasets.

Area of Science:

• Machine Learning
• Medical Data Analysis
• Computer Science

Background:

• Medical datasets frequently exhibit severe class imbalance, with normal samples vastly outnumbering abnormal ones.
• Traditional classification methods struggle with imbalanced data, often misclassifying minority (abnormal) classes as majority (normal).

Purpose of the Study:

• To develop a robust method for accurately identifying abnormal cases within imbalanced medical datasets.
• To propose a novel ensemble technique, the ensemble of rotation trees (ERT), specifically designed for imbalanced medical data.

Main Methods:

• The ensemble of rotation trees (ERT) method involves undersampling normal class data.
• Balanced training sets are created by combining undersampled normal data with abnormal data.
• Rotation matrices are applied to enhance diversity among ensemble members, followed by decision tree learning.

Main Results:

• Experimental results demonstrate that ERT significantly outperforms existing state-of-the-art methods on imbalanced medical datasets.
• The proposed undersampling and rotation matrix techniques improve model performance on the minority abnormal class.

Conclusions:

• The ensemble of rotation trees (ERT) offers a simple yet effective solution for handling class imbalance in medical datasets.
• ERT enhances the accurate detection of abnormal medical conditions, a critical task in clinical practice.