A classification for complex imbalanced data in disease screening and early diagnosis.

Area of Science:

• Statistics and Machine Learning
• Biomedical Data Science

Background:

• Imbalanced classification is challenging, especially with high-dimensional longitudinal health data.
• Traditional methods struggle with skewed data distributions common in disease diagnosis.
• Big data in health research exacerbates difficulties in imbalanced classification.

Purpose of the Study:

• To propose a novel nonparametric classification approach for imbalanced data in longitudinal and high-dimensional settings.
• To enhance disease diagnosis accuracy and provide interpretable feature selection.
• To reduce computational complexity in analyzing complex health datasets.

Main Methods:

• Functional principal component analysis (FPCA) for feature extraction from longitudinal data.
• Univariate exponential loss function with group LASSO penalty for high-dimensional classification.
• Nonparametric classification framework tailored for imbalanced and complex data structures.

Main Results:

• Significant improvement in imbalanced classification performance.
• Meaningful feature selection for enhanced interpretability.
• Remarkably lower computational complexity compared to existing methods.
• Successful application in early detection of Alzheimer's disease.

Conclusions:

• The proposed method effectively addresses imbalanced classification challenges in high-dimensional longitudinal data.
• It offers a powerful tool for disease diagnosis, exemplified by Alzheimer's detection.
• The approach balances accuracy, interpretability, and computational efficiency.