A Hybrid Clustering and SVD-Based Approach for Fuzzy-Neural System Modeling

We propose a fuzzy-neural modeling approach for automatically constructing a fuzzy-neural model from a set of input-output data. The proposed approach consists of two phases, structure identification and parameter identification. In the structure identification phase, rough TSK fuzzy rules are extracted through a clustering algorithm. Then a fuzzy neural network is built in the parameter identification phase for refining the parameters in each rule. To increase the learning speed, we adopt a hybrid learning which is a combination of recursive SVD-based least squares estimator and gradient descent method. Experimental results have shown that our proposed method is able to achieve lower approximation error and faster convergence rate.