Abstract
We introduce the Qualitative Sensitivity Analysis (QSA) technique and its application to generate automatically diagnostic rules for systems modeled in rule based causal format. The rule set is composed of atomic propositions. First, we introduce the Modified Causal Ordering (MCO) technique to derive ordering among the propositions. The result of MCO are complete subsets of propositions and Ordinal Reachability Graph (ORG). Each complete subset is identified by its ranking code. In QSA, propositions stand for the landmarks of the qualitative variables. QSA begins with the ORG and replaces the propositions with their equivalent landmarks. Perturbation is modeled by forcing a landmark to shift to its neighboring landmark. Sensitivity is identified as a consequence of perturbation, forcing two or more of the landmarks of a variable having the same rank. QSA can derive the sensitivity of the higher rank landmark due to perturbation affecting the lower rank ones. Applicability of the method is demonstrated through an example.
