Sensitivity analysis in hierarchical fuzzy logic models

A method developed previously for performing sensitivity analysis with systems based on two-valued logic is extended to the case in which all points in the closed unit interval are possible truth values and the rules of fuzzy logic apply. The advantage of this approach over incrementing inputs and observing outputs are that (1) some insight may be gained into the structure of the model, as when one node 'blocks' changes in another node from propagating upward, and (2) the sensitivity measures are proved to be true (one-sided) derivatives, which may not be the case when simulations are performed with positive and negative increments. These results apply to an important class of information structures found in knowledge-based decision support systems.<>