Genetic algorithm optimization of knowledge extraction from neural networks

Neural networks have been criticized for their lack of human comprehensibility. First, this paper proposes an extraction method of crisp if-then rules from ordinary backpropagation neural networks. Then, the paper presents a mechanism that compiles a neural network into an equivalent set of fuzzy rules. Genetic algorithms are used to find the correct structure of the fuzzy model that is equivalent to the neural network, and then to find the best shape of the membership functions. In order to reduce the number of fuzzy rules when we wish to compile a neural network with many inputs, genetic algorithms are used to find the best hierarchical structure of the fuzzy rules, considering the relations between the network inputs.