Total interaction index: A variance-based sensitivity index for second-order interaction screening

摘要: Abstract Sensitivity analysis aims at highlighting the input variables that have significant impact on a given model response of interest. By analogy with total sensitivity index, used to detect most influential variables, screening interactions can be done efficiently so-called interaction index (TII), defined as superset importance pair variables. Our aim is investigate TII, focus statistical inference. At theoretical level, we derive its connection and closed indices. We present several estimation methods prove asymptotical efficiency Liu Owen estimator. also address question estimating full set TIIs, budget function evaluations. observe pick-and-freeze method TIIs estimated linear cost respect problem dimension. The different estimators are then compared empirically. Finally, an application aiming discovering block-additive structure function, where no prior knowledge available, neither about nor blocks.