A generalized l _2,p-norm regression based feature selection algorithm.

Area of Science:

• Data Science
• Machine Learning
• Computational Biology

Background:

• Feature selection is crucial for high-dimensional data reduction in areas like genetics and image processing.
• Existing methods often rigidly use the same matrix norm, limiting applicability and performance.
• Current algorithms can be computationally expensive or provide suboptimal results due to approximations.

Purpose of the Study:

• To propose a generalized norm regression based feature selection (RFS) method.
• To extend existing optimization criteria by allowing different matrix norms in loss and regularization.
• To develop an efficient and robust feature selection technique for high-dimensional datasets.

Main Methods:

• Introduced a generalized norm regression framework with a novel optimization criterion.
• Allowed for different matrix norms in the loss function and regularization terms.
• Employed an iterative re-weighted least squares (IRLS) procedure with the LSQR algorithm for efficient solutions.

Main Results:

• The proposed generalized RFS method demonstrated robust performance on gene expression and image datasets.
• The approach effectively handles high-dimensional data, outperforming existing feature selection techniques.
• The iterative re-weighted least squares procedure with LSQR provided efficient computation.

Conclusions:

• The generalized norm regression feature selection offers a flexible and efficient solution for high-dimensional data.
• This method overcomes limitations of previous approaches by accommodating varying matrix norms.
• The proposed algorithm is suitable for practical applications in genetic data analysis and image processing.