This paper proposes a novel gradient-based rule learning system, the Fuzzy Rule-based Reasoner (FRR), to address optimization and scalability challenges while maintaining the transparency and interpretability of rule-based models. Unlike existing neurofuzzy approaches, FRR maximizes interpretability by utilizing semantically meaningful fuzzy logic segmentation and sufficient (single-rule) decision-making. Through extensive evaluations on 40 datasets, FRR demonstrates superior performance over existing rule-based methods (with an average accuracy improvement of 5% over RIPPER), comparable accuracy to tree-based models (using a 90% smaller rule base), and achieving 96% of the accuracy of state-of-the-art additive rule-based models (using only 3% of the rule base size).
