Dual regularized subspace learning using adaptive graph learning and rank constraint: Unsupervised feature selection...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-dimensional data, common in gene expression, presents challenges like the curse of dimensionality.
• Microarray datasets often have more features than samples, leading to ill-posed problems.
• Redundant features degrade learning algorithm performance and increase computational time.

Purpose of the Study:

• To develop a robust unsupervised feature selection method for unlabeled data.
• To address the curse of dimensionality in gene expression analysis.
• To select a discriminative feature subset by maximizing relevancy and minimizing redundancy.

Main Methods:

• Nonnegative Matrix Factorization (NMF) for data decomposition.
• Dual regularization using inner product norm for feature and representation matrices.
• Adaptive structure learning to preserve local information.
• Rank constraint using Schatten-p norm.

Main Results:

• The proposed method effectively discards redundant features while retaining informative ones.
• Demonstrated superior performance on six benchmark microarray datasets.
• Outperformed eight state-of-the-art unsupervised feature selection techniques.

Conclusions:

• The novel robust unsupervised feature selection technique is effective for high-dimensional gene data.
• Achieved improved clustering accuracy and normalized mutual information.
• Offers a valuable approach for analyzing unlabeled biological data.