Feature Selection in High-Dimensional Models via EBIC with Energy Distance Correlation.

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• High-dimensional models present challenges for traditional feature selection methods.
• Ordinary correlation coefficients only detect linear associations, limiting their effectiveness.
• Existing methods like Luo and Chen's may lack sufficient power in complex scenarios.

Purpose of the Study:

• To develop a more powerful feature selection method for high-dimensional data.
• To introduce energy distance correlation as a superior measure of variable dependence.
• To utilize extended Bayesian information criteria (EBIC) as an effective stopping criterion.

Main Methods:

• Proposed a novel feature selection algorithm incorporating energy distance correlation.
• Employed extended Bayesian information criteria (EBIC) for model selection and stopping criteria.
• Compared the proposed method against Luo and Chen's method using simulations and a real-world dataset.

Main Results:

• The new method demonstrates superior power in feature selection compared to Luo and Chen's approach.
• Energy distance correlation effectively captures both linear and non-linear associations.
• The proposed algorithm is proven to be selection-consistent.

Conclusions:

• The integration of energy distance correlation and EBIC offers a robust feature selection strategy.
• This enhanced method provides a more comprehensive analysis of variable relationships in high-dimensional settings.
• The findings are validated through empirical evidence and theoretical consistency proofs.