Robust hypergraph regularized non-negative matrix factorization for sample clustering and feature selection in...

Area of Science:

• Machine Learning
• Data Mining
• Computational Biology

Background:

• Non-negative matrix factorization (NMF) is a popular data representation technique for clustering and feature selection.
• Existing NMF methods often fail to capture intricate data geometry and are susceptible to noise and outliers.

Purpose of the Study:

• To develop a novel NMF framework, robust hypergraph regularized non-negative matrix factorization (RHNMF), addressing limitations of previous methods.
• To enhance the exploration of hidden geometrical structures within data.
• To improve the model's resilience against noise and outliers.

Main Methods:

• Introduced RHNMF, incorporating hypergraph Laplacian regularization to capture high-order data relationships.
• Utilized L2,1-norm constraint for residual estimation to enhance robustness against noise and outliers.
• Applied the RHNMF model to clustering and common abnormal expression gene selection tasks.

Main Results:

• RHNMF effectively captures geometric information and high-order data relationships.
• The L2,1-norm constraint significantly improves robustness to noise and outliers.
• Experimental validation on multi-view datasets demonstrates superior performance compared to state-of-the-art methods.

Conclusions:

• The proposed RHNMF model offers a robust and effective approach for data representation, clustering, and feature selection.
• RHNMF's ability to leverage hypergraph structures and handle noisy data makes it a valuable tool in data analysis.
• The model shows significant improvements over existing NMF-based techniques in various applications.