Component elimination strategies to fit mixtures of multiple scale distributions

Florence Forbes (1) , Alexis Arnaud (1) , Benjamin Lemasson (2) , Emmanuel Barbier (2)

1 MISTIS - Modelling and Inference of Complex and Structured Stochastic Systems
2 GIN - Grenoble Institut des Neurosciences

We address the issue of selecting automatically the number of components in mixture models with non-Gaussian components. As a more efficient alternative to the traditional comparison of several model scores in a range, we consider procedures based on a single run of the inference scheme. Starting from an overfitting mixture in a Bayesian setting, we investigate two strategies to eliminate superfluous components. We implement these strategies for mixtures of multiple scale distributions which exhibit a variety of shapes not necessarily elliptical while remaining analytical and tractable in multiple dimensions. A Bayesian formulation and a tractable inference procedure based on variational approximation are proposed. Preliminary results on simulated and real data show promising performance in terms of model selection and computational time.

Domaines: Statistiques [math.ST]

Florence Forbes, Alexis Arnaud, Benjamin Lemasson, Emmanuel Barbier. Component elimination strategies to fit mixtures of multiple scale distributions. RSSDS 2019 - Research School on Statistics and Data Science, Jul 2019, Melbourne, Australia. pp.81-95, ⟨10.1007/978-981-15-1960-4_6⟩. ⟨hal-02415090⟩