A Two-Component G-Prior for Variable Selection

Area of Science:

• Statistics
• Bayesian inference
• Machine learning

Background:

• Variable selection is crucial in statistical modeling for identifying relevant predictors.
• Zellner's g-prior is a common Bayesian approach for variable selection in linear models.
• Existing methods may lack efficiency or optimal performance in certain scenarios.

Purpose of the Study:

• To develop a more efficient Bayesian variable selection method.
• To introduce a novel two-component G-prior for enhanced performance.
• To demonstrate the method's effectiveness on real-world data.

Main Methods:

• A Bayesian variable selection approach is proposed.
• The method extends Zellner's g-prior by introducing a two-component G-prior.
• A tuning parameter is incorporated and calibrated using pseudo-variables.

Main Results:

• The proposed two-component G-prior demonstrates superior efficiency over Zellner's g-prior.
• Simulation studies show that models selected by the new method have smaller losses.
• The method performs favorably compared to other considered variable selection techniques.

Conclusions:

• The proposed two-component G-prior offers an efficient and effective Bayesian variable selection strategy.
• The method is robust and performs well on gene expression and environmental data.
• This approach provides a valuable tool for statistical modeling and data analysis.