A Genetically Optimized Method for Weight Updating in Fuzzy Cognitive Maps

Fuzzy Cognitive Map (FCM) offers a flexible modeling environment for handling both uncertainty and cognitive mapping. This recurrent structure combines the generalization capacity of fuzzy and learning capacity of neural networks for decision making. In traditional FCM, the weight matrix showing relationship between concepts in a particular domain was developed by experts. Correctly computed weight matrix will shows the intelligent behavior of a system by letting the relevant concepts to interact each other there by producing the required output using an inference rule. In situations where no experts are available, learning from existing data is the only choice. Weight matrix learning can be formulated as a heuristic search problem that finds suitable weight combinations. Instead of simply generating a random matrix from available historic data, various similarity measures were combined together with the optimization capacity of genetic algorithms for producing the weight matrix. The proposed method focus on reducing the difference between obtained response and actual response in minimum number of iterations.