Building predictive models from IF-THEN rules extracted from Fuzzy Cognitive Maps

In previous work, extraction algorithms have been used to obtain sets of causal IF-THEN rules that summarise and explain the dominant dynamics of systems modelled by means of Fuzzy Cognitive Maps. This article explains how to turn those rules into predictive models that respond in real time to the state of a set of input variables and produce an output predicting the future state of the system, either as a directional signal, as a probability distribution or as both. The article discusses 1) the fundamental design decisions —separating causal variables from target variables, input modes, normalisation of the evidence—, and 2) the solutions to the most frequent problems —uneven coverage across concepts, secondary concepts, domain constraints— and includes a final section devoted to the use of large language models (LLMs) as assistants for implementing these models in an automated way.