Regularized extreme learning adaptive neuro-fuzzy algorithm for regression and classification

This paper incorporates the regularization strategy of kernel based extreme learning machines (ELM) to improve the performance of a neuro-fuzzy learning machine. The proposed learning machine, regularized extreme learning adaptive neuro-fuzzy inference system (R-ELANFIS), has the advantages of reduced randomness, reduced computational complexity and better generalization. The parameters of the fuzzy layer of R-ELANFIS are randomly selected by incorporating the explicit knowledge representation using fuzzy membership functions. The parameters of the linear neural layer are determined by solving a constrained optimization problem in a regularized framework. Simulations on regression problems show that R-ELANFIS achieves similar or better generalization performance compared to well known kernel based regression methods and ELM based neuro-fuzzy systems. The proposed method can also be applied to multi-class classification problems.