A New Approach to Parameters Identification of Fuzzy Regression Models

We present a new approach to parameters identification of the fuzzy regression model with respect to the e-insensitive estimator in this paper. The proposed method firstly employs the improved fuzzy c-mean clustering algorithm to carry out fuzzy partition of input-output data pairs, which ascertains the membership functions of fuzzy system. Secondly, the quadratic convex optimization similar to the optimization in support vector regression machine is obtained based on e-insensitive estimator, which guarantees the feasibility of parameters identification. Besides, a comparison between the fuzzy regression system based on the e-insensitive estimator and that based on the least square estimator is made according to the performance index of root mean square error. The results show that the fuzzy regression models based on the proposed method are more insensitive to a small number of outliers and the number of clusters than that based on the least squares estimator.