Fuzzy Cognitive Maps (FCMs) are a very peculiar type of recurrent neural

networks that allow modeling complex systems in terms of concepts and causal relations.

While FCMs have proven successful in addressing simulation scenarios, their

performance in solving pattern classification problems is quite variable. Nevertheless,

the interpretability and transparency attached to these cognitive networks have motivated

researchers to improve their performance. On the other hand, some definitions and

theorems have been recently introduced to unveil the dynamic behavior in FCMs

equipped with sigmoid transfer functions. These analytic tools allow estimating bounds

for the activation value of each neuron. In this paper, we present a new learning

methodology for FCM-based classifiers, using the estimated bounds of neurons, that

leads to high prediction rates. The numerical results show that our neural classifier is

able to outperform most algorithms adopted for comparison, while remaining competitive

with the most accurate ones.