Abstract:
We consider the optimistic score ratio for robust Bayesian classification when the class-conditional distribution of the features is not perfectly known. The optimistic score searches for the distribution that is most plausible to explain the observed test sample among all distributions belonging to the class-dependent ambiguity set which is prescribed using a moment-based divergence. We show that the classification approach using optimistic score ratio is conceptually attractive, delivers rigorous statistical guarantees and is computationally tractable. We showcase the power of the proposed optimistic score ratio classifier on both synthetic and empirical data.
