Complexity Issues in Data-Driven Fuzzy Inference Systems: Systematic Literature Review

Abstract

The development of a data-driven fuzzy inference system (FIS) involves the automatic generation of membership functions and fuzzy if-then rules and choosing a particular defuzzification approach. The literature presents different techniques for automatic FIS development and highlights different challenges and issues of its automatic development because of its complexity. However, those complexity issues are not investigated sufficiently in a comprehensive way. Therefore, in this paper, we present a systematic literature review (SLR) of journal and conference papers on the topic of FIS complexity issues. We review 1 340 papers published between 1991 and 2019, systematize and classify them into categories according to the complexity issues. The results show that FIS complexity issues are classified as follows: computational complexity, fuzzy rules complexity, membership functions complexity, input data complexity, complexity of fuzzy rules interpretability, knowledge inferencing complexity and representation complexity, accuracy and interpretability complexity. The results of this study can help researchers and practitioners become familiar with existing FIS complexity issues, the extent of a particular complexity issue and to decide for future development.