IT2CFNN: An interval type-2 correlation-aware fuzzy neural network to construct non-separable fuzzy rules with uncertain and adaptive shapes for nonlinear function approximation

In this paper, a new interval type-2 fuzzy neural network able to construct non-separable rules with adaptive shapes is introduced. To reflect the uncertainty, shape of sets considered be uncertain. Therefore, form based on general Gaussian model different (including triangular, bell-shaped, trapezoidal) proposed. consider interactions among input variables, vectors are transformed feature spaces uncorrelated variables proper for defining each rule. Next, features fed fuzzification layer using proposed shape. Consequently, shapes, considering local and uncertainty formed. For type reduction contribution upper lower firing strengths rule adaptively selected separately. train parameters network, Levenberg-Marquadt optimization method utilized. The performance investigated clean noisy datasets show ability uncertainty. Moreover, paradigm, successfully applied real-world time-series predictions, regression problems, nonlinear system identification. According experimental results, our outperforms other methods more parsimonious structure.