A general adaptive unsupervised feature selection with auto-weighting.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• High-dimensional data presents challenges in efficiency and reliability.
• Unsupervised feature selection (UFS) is crucial due to the cost of data labeling.
• Existing embedded UFS methods often struggle with controlling sparsity and preserving feature structure.

Purpose of the Study:

• To propose a novel unsupervised feature selection model, GAWFS.
• To address limitations of existing UFS methods, particularly those using sparse projection matrices.
• To identify features that enhance data clustering without altering the original feature space.

Main Methods:

• Developed a General Adaptive Unsupervised Feature Selection with Auto-weighting (GAWFS) model.
• Employed non-negative matrix factorization and adaptive graph learning.
• Utilized a feature weighting matrix (Θ) to identify discriminative features and perform feature selection.

Main Results:

• GAWFS effectively identifies discriminative features for clustering.
• The method avoids projecting data into a low-dimensional embedding space, preserving original feature structure.
• Experimental results show GAWFS outperforms several state-of-the-art UFS methods on synthetic and real-world datasets.

Conclusions:

• GAWFS offers a superior approach to unsupervised feature selection.
• The model's auto-weighting mechanism provides effective feature filtering.
• GAWFS is a promising technique for efficient and reliable high-dimensional data analysis.