Global Sensitivity Analysis and Wasserstein Spaces

Sensitivity indices are commonly used to quantify the relative influence of any specific group input variables on output a computer code. In this paper, we focus both codes, which is cumulative distribution function, and stochastic codes. We propose way perform global sensitivity analysis for these kinds first setting, define two indices: one based Wasserstein Fréchet means, while second Hoeffding decomposition indicators balls. Further, when dealing with an “ideal version” code that fits into framework setting. Finally, deduce procedure realize level analysis, namely, interested in related distributions rather than inputs themselves. Several numerical studies proposed as illustrations different settings.