Conceptual

Generalized Heterogeneous Functional Model for Data-Driven Subgroup Discovery

A generalized functional regression method that simultaneously estimates subject-specific functional coefficient curves and identifies latent subgroups by fusing similar coefficient functions with a pairwise fusion penalty, without pre-specifying the number of subgroups. A pre-clustering step produces a finer-than-truth partition of subjects to reduce parameters and make estimation scalable to large mobile-health datasets, demonstrated on UK Biobank wearable activity data for mental-disorder and Parkinson's disease risk.