Meta-Fuzzy Items for Fuzzy Association Rules

Abstract

A large number of systems with a great predictive capacity, such as Deep Learning, are being currently used to solve a wide variety of real problems. However, the models obtained are not easy to understand by scientists, giving rise to the field of eXplainable Artificial Intelligence, which encourage techniques that obtain accurate and understandable models. Fuzzy Association Rules are models that can be understood by themselves, but its interpretability can be improved by representing the same information with fewer and simpler rules. In this work, we propose Meta-Fuzzy Items, which allows to define more generic fuzzy items to represent the same information with fewer rules, and to extend the type of associations that can be represented. Based on this proposal, a new fuzzy data-mining algorithm is presented to extract interesting and interpretable rules from quantitative transactions. The quality of our approach is analyzed using statistical analysis and comparing with a well-known fuzzy data-mining algorithm.