Explainable Fuzzy Modeling Through Antecedent Reduction and Antecedent Association in Type-1 ANFIS System

Abstract

The increasing demand for interpretable artificial intelligence (AI) has underscored the importance of explainable fuzzy models, particularly those based on Adaptive Neuro-Fuzzy Inference Systems. However, standard grid-based initialization often leads to exponential rule proliferation, significantly compromising model transparency and computational efficiency. To address this challenge, this study proposes an interpretable Type-1 fuzzy framework that utilizes Fuzzy C-Means clustering for data-driven membership function generation, followed by a systematic two-stage reduction strategy. The methodology integrates antecedent merging based on Euclidean distance to consolidate overlapping clusters and an activation-based pruning approach to eliminate inactive logic. Validated on the Fisher Iris dataset and Banknote Authentication dataset, the framework successfully reduced the rule base from 81 to 44 and 81 to 42 rules, achieving a 45.68% and 48.15% reduction in complexity, respectively. Crucially, this optimization enhanced classification accuracy from 76.00% to 93.33% for the Iris dataset and 44.08% to 51.13% for the Banknote Authentication dataset, demonstrating that removing redundant rules actively reduces logical noise. These results confirm that the proposed paradigm effectively balances parsimony with performance, offering a robust solution for explainable AI applications, with future extensions envisioned toward Type-2 systems for handling higher-order uncertainty.



Received: 30 November 2025 | Revised: 2 March 2026 | Accepted: 18 March 2026



Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.



Data Availability Statement

The data that support the findings of this study are openly available in the UCI Machine Learning Repository at https://archive.ics.uci.edu/dataset/53/iris and https://archive.ics.uci.edu/dataset/267/banknote+authentication.



Author Contribution Statement

Muhammad Hamza Azam: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Mohd Hilmi Hasan: Conceptualization, Resources, Supervision, Project administration, Funding acquisition. Saima Hassan: Writing – review & editing, Supervision. Noureen Talpur: Validation, Formal analysis, Writing – review & editing. Muhammad Huzaifa Azam: Writing – review & editing, Visualization.