Intelligent Rule Reduction for Improved ANFIS Performance in Classification

In data interpretation, particularly for low-dimensional data, fuzzy logic-based rule systems are valued for their transparency and interpretability in decision-making, crucial in sectors like healthcare and finance. However, these systems struggle with increasing dataset complexity and dimensionality. This study introduces a groundbreaking approach by integrating the Adaptive Neuro-Fuzzy Inference System (ANFIS) with Binary Particle Swarm Optimization (BPSO), marking the first instance where a metaheuristic technique is embedded within the ANFIS architecture for automated rule selection and reduction. The originality of our approach lies in the seamless synergy between the robustness of fuzzy logic and the dynamic optimization capability of BPSO as a specialized firing strengths selector meticulously tailored for effective rule reduction and performance optimization. The model minimizes rule updates and automatically adjusts BPSO parameters, aiming to improve interpretability and efficiency. Applied to eight classification benchmarks, the modified ANFIS demonstrates significant gains in computational efficiency, achieving time reductions as substantial as 90% to 98% across various datasets, and almost halving the number of rules compared to traditional ANFIS.