Many simulation models produce temporal or spatial data. Sensitivity analyses (SA) of such models require the use of the so-called Multivariate Sensitivity Analysis (MSA) methods that are often based on a dimension reduction principle. Model outputs are typically projected onto predefined or data-driven orthogonal bases such as polynomial or eigen vectors [1,2]. Classical scalar sensitivity analysis methods applied on the coefficients of the different components may give some rich information on the impact of model input factors on the variation of the shape of dynamic or spatial model outputs. Generalized Sensitivity Indices (GSIs) ([2,3]) can also be used to rank the overall contribution of input factors on the variability of these model outputs. The use of projection-based methods is however so far limited by the selection of the associated bases which is constrained by orthogonality requirements. Indeed, these bases does not always allow to extract relevant and interpretable information on structural properties of multivariate outputs. More applicable MSA methods are thus expected to be developed [4]. Clustering techniques have been designed to identify groups of similar objects in multivariate data sets. They may thus be particularly adapted to capture the variability of behaviors of model multivariate outputs. While binary clustering has been extensively used in scalar SA to assess the importance of factors leading to region of interest [5], there is still a need of quantitative sensitivity analysis methods taking benefit of multivariate outputs clustering with any number of clusters. In this work, we propose to make use of a fuzzy clustering procedure to enhance results of MSA on model with multivariate outputs. The main idea relies on the extensive use of the output of a fuzzy clustering method: the so-called membership functions (MF, valued in [0,1]), which quantify for any model response the degree of membership to each cluster. Membership function exactly correspond to posterior probability of membership produced by mixture-based clustering methods. Our approach is based on the analogy between MF and classical MSA projections. Indeed, MF provide a kind of decomposition, as they sum up to one. However, they are different to basis decomposition as they do not provide an additive decomposition among orthogonal components. Nevertheless, this analogy allows to extend MSA indices to clustered outputs. More precisely, we introduce: - MF-Sensitivity Indices (MF-SI) : SI on a MF of a given cluster. They allow to answer the question ‘which parameters influence the membership to a given cluster?’ - dMF-Sensitivity Indices (dMF-SI) : SI on pair-wise MF differences. They allow to answer the question ‘which parameters drive the output from one cluster to another? - MF-Generalized sensitivity indices (MF-GSI) : GSI computed on a MF vector to answer the question ‘‘What is the overall contribution of the parameters wrt change of behavior / clusters.’ We present the computation of these indices on a dedicated toy model producing temporal signals with one or two maxima in response to five parameters and show that the model behavior can be efficiently reported by the newly proposed indices. References: [1] Campbell, K., McKay, M. D., & Williams, B. J. (2006). Sensitivity analysis when model outputs are functions. Reliability Engineering & System Safety, 91(10-11), 1468-1472. [2] Lamboni, M., Monod, H., & Makowski, D. (2011). Multivariate sensitivity analysis to measure global contribution of input factors in dynamic models. Reliability Engineering & System Safety, 96(4), 450-459. [3] Lamboni M., (2018). Multivariate sensitivity analysis: minimum variance unbiased estimators of the first-order and total-effect covariance matrices. Reliability Engineering & System Safety 06/2018;, DOI:10.1016/j.ress.2018.06.004. [4] P. Wei, Z. Lu, J. Song, Variable importance analysis: a comprehensive review, Reliability Engineering & System Safety 142 (2015) 399–432. [5] Raguet, H., & Marrel, A. (2018). Target and Conditional Sensitivity Analysis with Emphasis on Dependence Measures. arXiv preprint arXiv:1801.10047.