Nearest-Neighbor Spline Approximation (NNSA) Improvement to TSK Fuzzy Systems

In this paper, we propose two versions of an improved defuzzification technique for Takagi Sugeno Kang (TSK) fuzzy systems (FSs) based on local third-order approximations. The presented nearest-neighbor spline approximation algorithms (NNSA1 and NNSA2) use the concept of a zeroth-order TSK FS and produce smooth surfaces with increased accuracy. The proposed methods are tested on a variety of function approximation problems pertaining to industrial applications against popular machine learning methodologies. Experimental results show that the proposed methods are indeed competitive in terms of computation time, approximation accuracy, and generalization ability when compared with other popular approaches.