Designing Highly Interpretable Fuzzy Rule-Based Systems with Integration of Expert and Induced Knowledge

This work describes a new methodology for fuzzy system modeling focused on maximizing the interpretability while keeping high accuracy. In order to get a good interpretability-accuracy trade-off, it considers the combination of both expert knowledge and knowledge extracted from data. Both types of knowledge are represented using the fuzzy logic formalism, in the form of linguistic variables and rules. The integration process is made carefully at both levels variables and rules, avoiding contradictions and/or redundancies. Results obtained in a well-known benchmark classification problem show the methodology ability to generate highly interpretable knowledge bases with a good accuracy, comparable to that achieved by other methods.