Evolving Fuzzy and Neuro-Fuzzy Systems: Fundamentals, Stability, Explainability, Useability, and Applications

This chapter provides an all-round picture of the development and advances in the fields of evolving fuzzy systems (EFS) and evolving neuro-fuzzy systems (ENFS) which have been made during the last 20 years since their first-time appearance around the year 2000. Their basic difference to the conventional (neuro-)fuzzy systems is that they can be learned, from data, on-the-fly during (fast) online processes in an incremental and mostly single-pass manner. Therefore, they stand for emerging topic in the field of soft computing and artificial intelligence for addressing modeling problems in the quickly increasing complexity of online data streaming applications and Big Data challenges, implying a shift from batch off-line model design phases (as conducted since the 1980s) to online (active) model teaching and adaptation. The focus will be on the definition of various model architectures used in the context of EFS and ENFS, on providing an overview of the basic learning concepts and learning steps in E(N)FS with a wide scope of references to E(N)FS approaches published so far (fundamentals), and on discussing advanced aspects towards an improved stability, reliability, and useability (usually must-to-haves to guarantee robustness and user-friendliness) as well as towards an educated explainability and interpretability of the models and their outputs (usually a nice-to-have to find reasons for predictions and to offer insights into the systems’ nature). The chapter will be concluded with a list of real-world applications where various E(N)FS approaches have been successfully applied with satisfactory accuracy, robustness, and speed.