Mixture of linear experts model for censored data: A novel approach with scale-mixture of normal distributions

Abstract Mixture of linear experts (MoE) model is one the widespread statistical frameworks for modeling, classification, and clustering data. Built on normality assumption error terms mathematical computational convenience, classical MoE has two challenges: (1) it sensitive to atypical observations outliers, (2) might produce misleading inferential results censored The aim then resolve these challenges, simultaneously, by proposing a robust model-based discriminant data with scale-mixture normal (SMN) class distributions unobserved terms. An analytical expectation–maximization (EM) type algorithm developed in order obtain maximum likelihood parameter estimates. Simulation studies are carried out examine performance, effectiveness, robustness proposed methodology. Finally, real dataset used illustrate superiority new model.