A Review on the Interpretability of Fuzzy Rule-Based Models in Modern Machine Learning

Interpretability in machine learning (ML) has emerged as a critical requirement for ensuring trust, safety, and ethical accountability, particularly in high-stakes domains such as healthcare, finance, and autonomous systems. Fuzzy rule-based models (FRBMs) are inherently interpretable due to their use of human-readable “if-then” rules, yet they are underrepresented in contemporary discussions of interpretable ML. This review critically examines FRBMs through three lenses: scale (local vs. global interpretability), audience (model designers, domain experts, end users), and model stage (inherent vs. post hoc interpretability). We analyze the conceptual foundations, methodologies, and evaluation techniques in both fuzzy modeling and ML literature. Key insights include the nuanced relationship between accuracy and interpretability, the role of constraint-based versus optimization-based modeling approaches, and the challenges of measuring semantic interpretability. Finally, we provide actionable recommendations for advancing FRBMs within modern interpretable ML, emphasizing audience-aware design, post hoc interpretability methods for complex models, and the development of standardized evaluation metrics. This review aims to situate FRBMs as a robust paradigm within explainable AI (XAI) and guide future research directions in interpretable ML.

Keywords:

Interpretability , Fuzzy Rule-Based Models , Explainable AI , Semantic Interpretability , High-Stakes Domains