.png){kind=icon}
Return
Tree-based heterogeneous block Gaussian graphical models
.png){kind=icon}
PRE
AI
DOI：10.1016/j.csda.2026.108399.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Gaussian graphical models reveal conditional dependence structures through precision matrix estimation. A novel estimator is developed for data with unknown sample heterogeneity and an unknown block structure among variables. In such a scenario, failure to account for heterogeneity fundamentally compromises analytical validity, and ignoring the block structure yields an overly complex graph that is difficult to interpret. The proposed solution integrates mixture modeling to capture heterogeneity, with the block structure induced by embedding tree-based information into the estimation process. A high-dimensional expectation-conditional-maximization algorithm is developed to implement the proposed method. Extensive simulations demonstrate its practical advantages over closely related alternatives. In a spectrometric data analysis, the proposed method is applied to food adulteration detection, successfully identifying heterogeneity and a block structure. Compared to existing methods, it provides more comprehensive and interpretable results.
Keywords:
Heterogeneity
Block graphical model
Network-based analysis
Mixture model
Regularization
Journal
C
IF:
1.6
Papers:
45
Citations:
0
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSave
ShareSave
View more