Variable selections in high-dimensional heterogeneous data

Abstract

This thesis explores statistical methods for structured variable selection in high-dimensional multi-response binary data. We investigate the integration of hierarchical structures into generalized linear mixed models through tree-guided regularization. An efficient optimization approach is developed and implemented with support for sparse computation. Simulation studies demonstrate the potential of this framework in recovering meaningful patterns and modeling trait correlations. The results contribute to the broader effort of scalable and interpretable multi-trait analysis.