Global sensitivity analysis of stochastic computer models with generalized additive models

The global sensitivity analysis, used to quantify the influence of uncertain input parameters on the response variability of a numerical model, is applicable to deterministic computer code (for which the same set of input parameters gives always the same output value). This paper proposes a new global sensitivity analysis method for stochastic computer code (having a variability induced by some uncontrollable parameters). The well-known framework of the joint modeling of the mean and dispersion of heteroscedastic data is used. To deal with the complexity of computer experiment outputs, a new non parametric joint model, based on two interlinked Generalized Additive Models (GAM), is proposed. The “mean model” allows to obtain the controllable parameters sensitivity indices, while the “dispersion model” allows to obtain the uncontrollable parameters ones. The relevance of this new model is analyzed with two case studies. Results show that the joint modeling approach leads accurate sensitivity index estimations even when clear heteroscedasticity is present.