A New Learning Approach for Fuzzy Cognitive Maps based on System Performance Indicators

Fuzzy cognitive map (FCM) is a simple and user friendly tool for modeling complex systems. It is described by the set of the concepts and the connections between them. FCM can be initialized based on expert knowledge or automatic constructed with the use of learning algorithms. Most learning methods focus on data error and the structure of the resulting model significantly differs from the reference object. This paper introduces a new multi-objective evolutionary approach for fuzzy cognitive maps learning based on system performance indicators. The proposed solution allows to select the most significant connections in terms of direct and indirect influence between concepts and obtain FCM models more similar to the reference structures. This approach was analyzed with the use of Elite Genetic Algorithm (EGA) and Individually Directional Evolutionary Algorithm (IDEA). Comparative analysis of the proposed methodology with the standard approach of FCMs learning was performed with the use of synthetic and real-life data. The obtained results show that the proposed approach based on system performance indicators outperforms the standard methodology focused only in data error.