Learning weighted linguistic fuzzy rules by using specifically-tailored hybrid estimation of distribution algorithms

Abstract

The WCOR methodology makes use of metaheuristic algorithms to find the best set of rules, as well as their weights, when learning weighted linguistic fuzzy systems from data. Although in early work based on this approach the search was carried out by means of a genetic algorithm, any other technique can be used. Estimation of distribution algorithms (EDAs) are a family of evolutionary algorithms in which the variation operator consists of a probability distribution that is learnt from the best individuals in a population and sampled to generate new ones. There are several possibilities for including problem domain knowledge in EDAs in order to make the search more efficient. In particular, this study examines specifically-designed EDAs which incorporate the information available about the WCOR problem into the probabilistic graphical model used to factorize the probability distribution. The experiments carried out with real and artificial datasets show an improvement in both the results obtained and the computational effort required by the search process.