On the meaning of sensitivity analysis

VIM establishes sensitivity as “the quotient of the change in an indication of a measuring instrument and the corresponding change in a given value of the quantity being measured”. In the era of digitalisation and AI, measurement models are becoming increasingly complex, with a high number of input parameters and deep nonlinearities. The sensitivity of a model evaluated according to ISO-GUM at one point can be considered a local parameter, valid only around the associated measurement point, which limits its interpretation and use. Two other methods aim at broadening the definition of sensitivity to improve its understanding and use for digital models: Global Sensitivity Analysis (GSA), which defines sensitivity as a global parameter, and Shapley value, which has been widely used for machine learning and AI models. This paper aims to understand these three approaches and to discuss the importance of sensitivity in the context of digital transformation.