Reliable Type 2 Fuzzy Min–Max neural networks for pattern classification

The Fuzzy Min–Max (FMM) algorithm is a powerful classification method capable of handling non-linear class boundaries and making both hard and soft decisions while learning from online data. However, it faces significant challenges, including sensitivity to the expansion coefficient, information loss during the contraction stage, and the overlap problem. To address these limitations, we propose a Reliable Type-2 Fuzzy Min–Max (RT2FMM) algorithm, which incorporates type-2 fuzzy logic to consider hyperbox uncertainty and effectively resolve the overlap problem. By assigning distinct certainties to overlapping regions, RT2FMM eliminates the need for the contraction stage and the overlap test. Additionally, we introduce weighted factors for hyperboxes, which enhances the reliability of membership values and models their mutual effects. Our comprehensive experimental evaluation across twenty datasets demonstrates that RT2FMM significantly outperforms existing FMM-based models in terms of robustness and accuracy. The Friedman test further confirms the superior performance of RT2FMM compared to commonly used classifiers, highlighting its potential as a robust solution for complex classification tasks. • Employ a type-2 fuzzy logic framework to handle the uncertainty of hyperboxes. • Introduce weights between the hyperbox and class layers to address the hyperbox’s reliability. • Eliminate the contraction stage and avoid losing learned information during the training stage. • Reduce sensitivity to the expansion coefficient hyperparameter. • Comprehensive evaluation and comparison on twenty UCI datasets.