Neuro-Fuzzy System with Adaptive Membership-Activation Functions for Pattern Recognition

This paper presents a novel neuro-fuzzy system designed for pattern recognition tasks, particularly classification. Building upon the Takagi-Sugeno-Kang model, our system improves performance by incorporating adaptive membership-activation functions that dynamically adjust during training, enhancing adaptability to varying data complexities. Additionally, it adopts a comprehensive learning approach, integrating lazy learning, self-learning, and controlled learning mechanisms to enhance classification capabilities in both batch and online learning scenarios. Despite its complexity, the system maintains computational simplicity, enabling rapid implementation and high-speed processing. Experimental results demonstrate its effectiveness in real-world pattern recognition tasks, achieving high-quality approximation even with limited training data.