LAUSR

This paper introduces the design methodology of a noise-robust fuzzy classifier based on type-2 fuzzy clustering and enhanced learning methods. The design procedure for the noise-robust fuzzy classifier (NrFC) can be divided into two parts. First, interval type-2 fuzzy c-means clustering is applied to the hidden layer to minimize the effect of noise or outliers when training the model. Second, an enhanced learning method is employed to train the connection weights between the hidden and output layers. The proposed NrFC uses a cross-entropy error function as its cost function. The Softmax function represents a categorical distribution located at the output layer nodes. In addition, the connection weights of the output layer are adjusted through nonlinear least squares-based learning, and L2 norm-regularization is considered to avoid the degradation of the generalization ability caused by overfitting. The learning mechanism is realized by adding the L2 penalty term to the cross-entropy error function. It is used to cope with over-fitting and multicollinearity problems, which generally appear in conventional fuzzy neural networks. The design methodology of the NrFC is discussed and analyzed using several publicly available benchmark datasets. The performance of the proposed networks is quantified through comprehensive experiments and comparative analysis.