Navigating interpretability issues in evolving fuzzy systems

Abstract

In this position paper, we are investigating interpretability issues in the context of evolving fuzzy systems (EFS). Current EFS approaches, developed during the last years, are basically providing methodologies for precise modeling tasks, i.e. relations and system dependencies implicitly contained in on-line data streams are modeled as accurately as possible. This is achieved by permanent dynamic updates and evolution of structural components. Little attention has been paid to the interpretable power of these evolved systems, which, however, originally was one fundamental strength of fuzzy models over other (data-driven) model architectures. This paper will present the (little) achievements already made in this direction, discuss new concepts and point out open issues for future research. Various well-known and important interpretability criteria will serve as basis for our investigations. © 2012 Springer-Verlag.