How choosing a useful statistical model ? Illustration with the mixture model

vendredi 24 février 2017, 11h00 - 12h00

Salle du conseil, espace Turing

The presentation will turn around the famous sentence of Box « All models are wrong, some are useful ». I will try to analyse the theoretical, the methodological and the practical impact of this sentence. To illustrate my comments, I will focus on the mixture of probability distributions which can be considered from several points of view (density estimation or model-based clustering for instance) and I will discuss the possible impact of these different views in the model selection perspective. I will tell some stories related to the Box citation which are not always success stories.