Knowledge Base Learning of Linguistic Fuzzy Rule-Based Systems in a Multi-objective Evolutionary Framework

We propose a multi-objective evolutionary algorithm to generate a set of fuzzy rule-based systems with different trade-offs between accuracy and complexity. The novelty of our approach resides in performing concurrently learning of rules and learning of the membership functions which define the meanings of the labels used in the rules. To this aim, we represent membership functions by the linguistic 2-tuple scheme, which allows the symbolic translation of a label by considering only one parameter, and adopt an appropriate two-variable chromosome coding. Results achieved by using a modified version of PAES on a real problem confirm the effectiveness of our approach in increasing the accuracy and decreasing the complexity of the solutions in the approximated Pareto front with respect to the single objective-based approach.