Dynamic Supervised Principal Component Analysis for Classification.

Area of Science:

• Machine Learning
• Statistical Analysis
• Data Science

Background:

• High-dimensional data classification faces challenges with evolving class distributions over time.
• Traditional discriminant analysis methods struggle with non-static and large datasets.
• Scalability and adaptability are critical for modern classification tasks.

Purpose of the Study:

• To introduce a novel framework for dynamic classification in high-dimensional spaces.
• To adapt discriminant analysis techniques for learning dynamic decision rules.
• To address the challenges of non-static class distributions and computational efficiency.

Main Methods:

• A new supervised dimension reduction method using kernel smoothing is proposed.
• The method is examined for its application in linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA).
• Numerical simulations and real-world data examples are used for evaluation.

Main Results:

• The proposed methods demonstrate considerable improvements in classification accuracy.
• Significant enhancements in computational efficiency were observed.
• The framework effectively handles evolving class distributions in high-dimensional data.

Conclusions:

• The developed framework offers a robust and adaptive solution for dynamic classification.
• The novel dimension reduction technique enhances performance in both LDA and QDA.
• This work advances the field by providing tools for non-static high-dimensional data analysis.