EXTENDED APPROACH FOR EVOLVING NEO-FUZZY NEURAL WITH ADAPTIVE FEATURE SELECTION

This paper introduces an evolving neo-fuzzy neural network with adaptive feature selection approach in which candidate models with larger and smaller number of input variables than the current model are developed concurrently. The best amongst the current and candidate models is chosen at each step. The approach uses an incremental learning algorithm to simultaneously update the weights, to select the input variables, and evolve the network structure. Computational experiments concerning identification of a nonlinear process is performed to evaluate the method and to compare its performance against alternative evolving models. The results show that the extended evolving neo-fuzzy neural network with adaptive feature selection approach achieves higher or as high performance as alternatives evolving modeling methods.