Estimate of quantile-oriented sensitivity indices

Abstract : In the context of black-box numerical codes, it is relevant to use sensitivity analysis in order to assess the influence of each random input X over the output Y. Goal-oriented sensitivity analysis states that one must first focus on a certain probability feature θ(Y) from the distribution of Y (such as its mean, quantile, or a probability of failure etc...), which would be chosen regarding a relevant strategy. The wish is to evaluate the impact of each input over θ(Y). In order to get supplementary information about sensitivity, we set that θ(Y) is the α-level quantile of Y , where α ∈]0, 1[. Throughout some examples, it has been pointed out that in some cases quantile-oriented sensitivity indices can detect some influence that Sobol indices would not. Mainly, the influence over each level of quantile displays how an input distribution entirely propagates through the output. We establish further results for the quantile-oriented indices properties in order to justify their relevancy. The main contribution of this paper comes when a statistical estimator for this index is introduced.
Document type :
Preprints, Working Papers, ...
Liste complète des métadonnées

Cited literature [13 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01450891
Contributor : Thomas Browne <>
Submitted on : Tuesday, January 31, 2017 - 2:32:15 PM
Last modification on : Friday, April 12, 2019 - 4:22:51 PM
Document(s) archivé(s) le : Monday, May 1, 2017 - 2:15:23 PM

File

Browne_submission.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01450891, version 1

Citation

Thomas Browne, Jean-Claude Fort, Bertrand Iooss, Loïc Le Gratiet. Estimate of quantile-oriented sensitivity indices. 2017. ⟨hal-01450891⟩

Share

Metrics

Record views

535

Files downloads

271