Fuzzy Cognitive Maps development using Ant Colony Optimization with local search procedure

Fuzzy Cognitive Maps (FCMs) can represent causal relations among concepts in the form of directed weighted graphs. Learning FCMs from data is a difficult task because of the large number of candidate FCMs. The goal of learning FCMs is to find good weights for the FCM so that the response of the learned FCM should be as close to the observed data as possible. The number of variables to be optimized grows quadratically to the number of concepts. Furthermore, the non-linear numerical optimization problem is also a NP-hard problem. A FCM development algorithm based on Ant Colony Optimization is proposed in this paper. A local search procedure is designed to improve the performance of the learning algorithm. The FCM learning problem is formulated as searching for weights of the FCM so that the FCM has a response sequence as close to the observed data as possible. Simulated data generated from several FCMs with different number of concepts is used to test the ACO-based learning algorithm. The proposed algorithm outperforms other data-driven FCM learning algorithms such as non-linear Hebbian learning, data-driven non-linear Hebbian learning and real-coded genetic algorithm in several experiment scenarios.