Analyzing and simplifying model uncertainty in fuzzy cognitive maps

Fuzzy Cognitive Mapping (FCM) represents the ‘mental model’ of individuals as a causal network equipped with an inference engine. As individuals may disagree or evidence be insufficient, causal links may be assigned a range rather than one value. When all links have range, the massive search space is a challenge to running simulations. In this paper, we presented, implemented, and evaluated a new approach to identify which ranges are important and simplify models accordingly. Our approach uses a factorial design of experiments, implemented using parallelism to offset its high computational cost. Our implementation (including our new Python library for FCM) is freely available on a third-party repository. Our evaluation on three previously published models shows that our approach can simplify almost half of a model under common settings, and runs within seconds on entry-level hardware for small FCMs. Further research is needed on simplifying the few FCMs having many links.