Principal Component Analysis Based on Graph Laplacian and Double Sparse Constraints for Feature Selection and Sample...

Area of Science:

• Data Science
• Machine Learning
• Bioinformatics

Background:

• Principal Component Analysis (PCA) is crucial for dimensionality reduction but often yields dense, hard-to-interpret components.
• Existing sparse PCA methods struggle with redundant principal components (PCs), limiting interpretability.
• Addressing these limitations is vital for accurate analysis of complex datasets.

Purpose of the Study:

• To propose a novel method, PCA based on Graph Laplacian and Double Sparse constraints (GDSPCA), for improved interpretation of principal components (PCs).
• To enhance PCA by considering the internal geometry of data and filtering redundant PCs.
• To achieve a more interpretable low-dimensional subspace representation.

Main Methods:

• GDSPCA integrates graph Laplacian to capture data's geometric structure.
• Simultaneous application of L2,1-norm and L1-norm regularization terms to enforce row and element sparsity, respectively.
• Filtering of redundant and irrelevant PCs to improve interpretability.

Main Results:

• GDSPCA effectively filters redundant and irrelevant PCs, leading to enhanced interpretability.
• The method successfully incorporates the geometric structure of the data into the PCA framework.
• Experimental results on multi-view biological data confirm the feasibility and effectiveness of GDSPCA.

Conclusions:

• GDSPCA offers a significant improvement over existing PCA and sparse PCA methods for data interpretation.
• The integration of graph Laplacian and double sparse constraints provides a robust approach to dimensionality reduction.
• GDSPCA demonstrates strong potential for analyzing complex, multi-view biological data.