The cluster graphical lasso for improved estimation of Gaussian graphical models

Area of Science:

• Statistics
• Machine Learning
• Network Analysis

Background:

• Estimating Gaussian graphical models is crucial in high-dimensional data analysis.
• The graphical lasso is a popular method, but its reliance on single linkage hierarchical clustering has limitations.

Purpose of the Study:

• To introduce a novel clustering-based approach for Gaussian graphical models.
• To address the limitations of single linkage clustering in the graphical lasso.

Main Methods:

• The study reveals a connection between the graphical lasso and single linkage hierarchical clustering.
• A new method, the cluster graphical lasso, is proposed, employing alternative clustering strategies before applying the graphical lasso within clusters.

Main Results:

• Model selection consistency is established for the cluster graphical lasso.
• The proposed method demonstrates superior performance compared to the standard graphical lasso in simulations and real-world data applications.

Conclusions:

• The cluster graphical lasso offers a more robust and accurate approach to Gaussian graphical model estimation.
• This technique improves network structure identification by utilizing more effective clustering methods.