B1777
Title: Flexible multivariate mixture models: A comprehensive approach for modeling mixtures of non-identical distributions
Authors: Samyajoy Pal - LMU Munich (Germany) [presenting]
Christian Heumann - Ludwig-Maximilians-University Munich (Germany)
Abstract: The mixture models are widely used to analyze data with cluster structures and the mixture of Gaussians is most common in practical applications. The use of mixtures involving other multivariate distributions, like the multivariate skew normal and multivariate general hyperbolic, is also found in the literature. However, in all such cases, only the mixtures of identical distributions are used to form a mixture model. A novel and versatile approach is presented for constructing mixture models involving identical and non-identical distributions combined in all conceivable permutations (e.g., a mixture of multivariate skew normal and multivariate general hyperbolic). Any conventional mixture model is also established as a distinctive particular case of the proposed framework. The practical efficacy of the model is shown through its application to both simulated and real-world data sets. The comprehensive and flexible model excels at recognizing inherent patterns and accurately estimating parameters.
