A genetic-based method for training fuzzy systems

In this paper, a genetic-based method for training fuzzy classification systems is proposed. The genetic algorithm, called genetic algorithm with no genetic operators (GANGO), neither needs to use the conventional genetic operators nor to store the population throughout the evolution process, but still has the same search mechanisms as conventional genetic algorithms. The novelty of the proposed training approach lies in: 1) the new scheme of encoding a fuzzy system based on the interpretation of the values of the components of a fuzzy relationship matrix as the sample probabilities of genes, and this, together with no requirement on storing the population, contributes to a dramatic decrease in storage requirement and computational cost; and 2) the automatic elimination of irrelevant fuzzy rules using a fitness reassignment strategy at the gene level and a weight truncation strategy. The proposed training method is successfully applied to train a fuzzy system for the classification of real-world remote sensing data