A Criterion for Local Model Selection

Abstract

In this paper, we introduce a class of local divergences between two probability distributions and illustrate its usefulness in model selection. Explicit expressions of the proposed local divergences are derived when the underlying distributions are members of the exponential family of distributions or they are described by multivariate normal models. In addition, a local model selection criterion, termed the local divergence information criterion (LDiv.IC), is proposed. Simulations and applications are presented in order to study and exemplify the performance of the proposed criterion.