Journal of Pioneering Artificial Intelligence Research

Abstract

Machine Learning and Fuzzy Logic effectively complement each other in neuro-fuzzy systems, combining data-driven learning with interpretable reasoning under uncertainty. A prominent architecture is the Adaptive Neuro-Fuzzy Inference System (ANFIS), which employs the Takagi-Sugeno method to model complex nonlinear systems. However, standard ANFIS is traditionally limited to the algebraic product operation in its antecedent layer and often encounters computational bottlenecks on CPU hardware when handling high-dimensional datasets. In this work, we propose a generalization of the ANFIS antecedent layer by systematically integrating various families of parametric and non-parametric t-norms. We evaluate this architecture in two distinct environments: a traditional CPU-based baseline and a GPU-accelerated framework. Experiments conducted on various benchmark datasets demonstrate that replacing the fixed product operator with flexible t-norms improves predictive accuracy and adaptability, and statistical tests confirm the benefits of the proposed generalization.