TS-fuzzy system-based support vector regression

Abstract

This paper proposes a new regression model, the Takagi–Sugeno Fuzzy System-based Support Vector Regression (TSFS-SVR). The TSFS-SVR is motivated by TS-type fuzzy rules and its parameters are learned by a combination of fuzzy clustering and linear SVR. In contrast to a kernel-based SVR, the TSFS-SVR has a smaller number of parameters while retaining the SVR's good generalization ability. In TSFS-SVR, a one-pass clustering algorithm clusters the input training data. A new TS-kernel, which corresponds to a TS-type fuzzy rule, is then constructed by the product of a cluster output and a linear combination of input variables. The TSFS-SVR output is a linear weighted sum of the TS-kernels. To achieve high generalization ability, TSFS-SVR weights are learned through linear SVR. This paper demonstrates the capabilities of TSFS-SVR by conducting simulations in clean and noisy function approximations and signal prediction. This paper also compares simulation results from the TSFS-SVR with Gaussian kernel-based SVR and other learning models.