An enhanced discriminability recurrent fuzzy neural network for temporal classification problems

This paper proposes an enhanced discriminability recurrent fuzzy neural network for temporal classification problems. To consider classification problems, the most important consideration is the ''discriminability''. To enhance the ''discriminability'', the feedback topology of the proposed fuzzy neural network is fully connected in order to handle temporal pattern behavior. Furthermore, the proposed fuzzy neural network considers minimum-classification-error and minimum-training-error. In minimum-classification-error, the weights are updated by maximizing the discrimination among different classes. In minimum-training-error, the parameter learning adopts the gradient descent method to reduce the cost function. Therefore, the novelty of the enhanced discriminability recurrent fuzzy neural network is that it not only minimizes the cost function but also maximizes the discriminability. It is constructed from structure and parameter learning. Simulations and comparisons with other recurrent fuzzy neural networks verify the performance of the enhanced discriminability recurrent fuzzy neural network under noisy conditions. Analysis results indicate that the proposed fuzzy neural network exhibits excellent classification performance.