Fuzzy Modeling Based on Ordinary Fuzzy Partitions and Nearest Neighbor Clustering

  • Authors:

  • George E. Tsekouras

  • Affiliations:

  • Department of Cultural Technology and Communication, University of the Aegean, Mytilene, Greece 81100

  • Venue:
  • Journal of Intelligent and Robotic Systems
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper a novel algorithm is proposed to train fuzzy models. The novelty of the contribution lies on the development of a nearest neighbor-clustering scheme, which is able to perform the structure identification and the model parameter estimation without taking into account any random initial guesses. This nearest neighbor-clustering search is based on defining ordinary fuzzy partitions in the input space, and produces a number of spherical-shaped fuzzy clusters. The number of these clusters provides the number of rules of the fuzzy model. The premise model parameters are obtained by projecting the spherical clusters on each axe. Relationally, the consequent model parameters are determined by applying the orthogonal least-squares algorithm. Finally, the above model parameters are fine tuned by using the gradient descent method. The whole scheme requires one-pass through the data set and therefore it is a fast procedure that is easy to implement. Simulation experiments verify the algorithm's efficiency with respect to its prediction performance, its initialization capabilities, and its speed.