Uncertainty and sensitivity analysis to complex systems

Abstract

In the complexity modeling, variance decomposition technique is widely used for the quantification of the variation in the output variables explained by covariates. In this work, the satisfaction of sampling-based variance decomposition strategy (SVDS) is firstly testified in the implementation of an analytic method for uncertainty and sensitivity analysis (UASA) of complex systems. Results suggest that SVDS may overvalue the impacts from individual covariates alone but underestimate the effects from their interactions when the model under discussion involves the interaction effects of nonlinear problems of individual covariates. Following the phenomenon, a modification of SVDS is proposed to generate sensitivity measures that well coincide with the analytic method. The testified strategy, together with our proposed modification, is then employed to clarify the roles of infectious rate and recovered rate, as well as of their interaction, in the estimation of equilibrium state (ES) for both SIR and SIS models. Results demonstrate that infectious and recovered rates almost play the same roles less crucial than that acted by the initial susceptible individuals in the decision of ES for SIR model, accompanied by a fragile contribution from their interactions; while in SIS model, infectious rate is more robust than recovered rate, and their interaction effect is also non-ignorable.