A0303
Title: Robust modeling of multivariate heterogeneous datasets using a tractable multivariate skew heavy-tailed distribution
Authors: Olcay Arslan - Ankara University (Turkey) [presenting]
Fatma Zehra Dogru - Giresun University (Turkey)
Abstract: Finite mixtures of multivariate normal distributions are often considered for modeling multivariate heterogeneous datasets. However, in applications, besides heterogeneity, datasets may have an asymmetric form with tail behavior different from the normal distribution so modeling them with a finite mixture of normal distributions may not provide an adequate model to represent all the features of the data. Therefore, recent research has focused on using finite mixtures of multivariate models with more flexible distributional forms to properly account for skewness, and heavy or light-tailedness to model multivariate heterogeneous data. As an alternative to the newly launched models in the literature, a finite mixture of multivariate skew Laplace normal (MSLN) distributions is introduced to simultaneously handle skewness and heavy tailedness in multivariate heterogeneous datasets. The MSLN distribution has recently been proposed with some plausible advantages over its counterparts. The proposed mixture model (FM-MSLN model) has some desirable properties, including a tractable density with a fewer number of parameters and ease of computation for simulation and estimation of parameters. Maximum likelihood parameter estimation of the FM-MSLN model via the expectation-maximization (EM) algorithm is given. The modeling performance of the FM-MSLN model is demonstrated using simulated datasets and a real data example.
