Group descent algorithms for nonconvex penalized linear and logistic regression models with grouped predictors

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• Penalized regression, including the lasso, is effective for variable selection.
• Many real-world datasets exhibit inherent grouping structures in variables.
• Selecting entire groups of variables is often more relevant than individual variable selection.

Purpose of the Study:

• To extend penalized regression frameworks for group selection.
• To introduce and evaluate group SCAD and group MCP methods.
• To develop stable and efficient algorithms for fitting these group selection models.

Main Methods:

• Extension of nonconvex penalties (SCAD, MCP) to group selection problems.
• Development of algorithms for stable and efficient fitting of group SCAD and group MCP models.
• Comparison of statistical properties through simulations and real data examples.

Main Results:

• The study presents algorithms for fitting group SCAD and group MCP models.
• Simulation results and real data examples are provided to compare these methods.
• The statistical properties of the group lasso, group SCAD, and group MCP are contrasted.

Conclusions:

• Group SCAD and group MCP offer viable alternatives for group selection in penalized regression.
• The developed algorithms enable stable and efficient application of these methods.
• The study provides insights into the comparative performance of different group selection techniques.