An importance quantification technique in uncertainty analysis for computer models

The authors have developed a technique to numerically quantify importance of input variables including uncertainties to the output uncertainty, based on the concept of uncertainty reduction, which makes it practically possible to estimate the importance measure proposed by Hora and Iman (1986).

Abstract

The authors have developed a technique to numerically quantify importance of input variables including uncertainties to the output uncertainty. The technique makes it practically possible to estimate the importance measure, proposed by Hora and Iman (1986), which is based on the concept of uncertainty reduction. The technique required a limited number of calculations based on the original model using the Monte Carlo or the Latin hypercube sampling. Effectiveness of the technique is demonstrated in a comparative study by applying the technique and a conventional regression method to two computer models, an analytical model and the TERFOC model.>