B0174
Title: Levels merging in the latent class model
Authors: Christophe Biernacki - Inria (France) [presenting]
Abstract: The latent class model (LCM), dedicated to cluster categorical variables, suffers from the curse of dimension when the number of levels is large, a situation frequently encountered in practice. It is proposed to extend LCM to natural modelling which limits the number of levels by merging them, a process which is also equivalent to a specific level clustering. Related estimation and model selection processes are also presented, discussed and numerically illustrated.