Mistura finita de modelos de regressão assimétricos com resposta censurada

Abstract

The process of modeling via linear regression to analyze data with censored response can become an arduous task when the data set presents certain particularities such as asymmetry, heavy tails and unobserved heterogeneity. Some models to examine data with these characteristics have already been proposed in the literature, however, none of them seeks to accommodate all three at the same time for censored data. The model presented in this work aims to meet this challenge by using the finite mixture of linear regressions in the context of the scale mixture of skew-normal (SMSN) class of distributions, achieving a model with high flexibility that will be useful in several areas of research. Due to the complexity of the likelihood function in these cases and the convenient hierarchical representation of the SMSN class of distributions, an EM type algorithm was developed to obtain the maximum likelihood estimators of the original parameters. To show the applicability of the model both in the context of fit data and in the classification context, simulation studies were performed and a real dataset was analyzed.