Proposing a global sensitivity analysis method for linear models in the presence of correlation between input variables

Sensitivity analysis is considered to be an important part of evaluating the performance of mathematical or numerical models. one-factor-at-a-time (oat) and dierential methods are among the most popular sensitivity analysis (sa) schemes employed in the literature. two major limitations of the above methods are lack of addressing the correlation between model factors and being a local method. given these limitations, its extensive use among modelers raises concern over the credibility of the associated sensitivity analyses. this paper proposes proof for the ineciency of the aforementioned methods drawn from experimental designs, and provides a novel technique based on principal component analysis (pca) to address the issue of the correlation between input factors. in addition, proper guidelines are suggested to handle other conditions.