1: \begin{abstract}

2: We study the properties of false discovery rate (FDR) thresholding,

3: viewed as a classification procedure. The ``$0$''-class (null) is

4: assumed to have a known density while the ``$1$''-class (alternative)

5: is obtained from the ``$0$''-class either by translation or by

6: scaling. Furthermore, the ``$1$''-class is assumed to have a small

7: number of elements w.r.t. the ``$0$''-class (sparsity). We focus on

8: densities of the Subbotin family, including Gaussian and Laplace

9: models. Nonasymptotic oracle inequalities are derived for the excess

10: risk of FDR thresholding. These inequalities lead to explicit rates of

11: convergence of the excess risk to zero, as the number $m$ of items to

12: be classified tends to infinity and in a regime where the power of the

13: Bayes rule is away from $0$ and $1$. Moreover, these theoretical

14: investigations suggest an explicit choice for the target level

15: $\alpha_m$ of FDR thresholding, as a function of $m$. Our oracle

16: inequalities show theoretically that the resulting FDR thresholding

17: adapts to the unknown sparsity regime contained in the data. This

18: property is illustrated with numerical experiments.

19: \end{abstract}

20: