FEATURE ELIMINATION IN KERNEL MACHINES IN MODERATELY HIGH DIMENSIONS.

Area of Science:

• Machine Learning
• Statistical Learning
• Kernel Methods

Background:

• Feature selection is crucial for model performance and interpretability in statistical learning.
• Kernel machines offer powerful non-linear modeling capabilities but can be sensitive to irrelevant features.

Purpose of the Study:

• To develop and analyze a novel recursive feature elimination approach for kernel machines.
• To establish the theoretical consistency of the proposed method in identifying the true feature space.
• To validate the approach through case studies and simulations.

Main Methods:

• Recursive feature elimination algorithm tailored for kernel machines.
• Theoretical analysis of the method's consistency properties under generalized assumptions.
• Empirical evaluation using case studies and simulation experiments.

Main Results:

• The proposed recursive feature elimination method is shown to be uniformly consistent.
• Theoretical guarantees for identifying the correct feature space are established.
• Practical applicability is demonstrated through case studies and simulations, showing strong performance.

Conclusions:

• The developed recursive feature elimination technique provides a theoretically sound and practically effective tool for feature selection in kernel machines.
• This method enhances model interpretability and performance by efficiently identifying relevant features.