Splitting K-means generated Neural Fuzzy System with Support Vector Regression

This paper proposes a Splitting K-means generated Neural Fuzzy System with Support Vector Regression (SKNFS SVR). The consequent layer in SKNFS-SVR is a Takagi-Sugeno-Kang (TSK)-type consequent. For structure learning, a splitting K-means algorithm clusters the input training data and determines the rule number. For parameter learning, a linear support vector regression (SVR) algorithm is proposed to tune free parameters in the consequent part. The motivation for using SVR for parameter learning is to improve the SKNFS-SVR generalization ability. This paper demonstrates the capabilities of SKNFS-SVR by conducting simulations in clean and noisy function approximations. This paper also compares simulation results from the SKNFS-SVR with Gaussian kernel-based SVR.