A novel uninorm-based evolving fuzzy neural networks

Abstract

Due to the constantly increasing rate in size and temporal availability of data, learning from data stream is a contemporary and demanding issue. This work presents a structure and introduces a learning approach to train an uninorm-based evolving fuzzy neural networks (UEFNN). The fuzzy rules can be stitched up or expelled by of statistical contributions of the fuzzy rules. The learning and modeling performances of the proposed UEFNN are validated using a benchmark problem. Simulation result and comparisons with state-of-art evolving neuro-fuzzy methods and demonstrate that our new method can compete and in some cases even outperform these approach in terms of RMSE and complexity.