Extracting fuzzy if-then rules by using the information matrix technique (Q1765218)

The authors suggest a new approach to extract fuzzy if-then rules from data including noise. The procedure does not depend on any pool of experts. The rules are extracted from an information matrix produced by unsupervised learning with the normal information diffusion. The main work of extracting rules is to calculate the centroids of the rows of an information matrix. With the extracted rules one has a fuzzy mapping to estimate a function. This new model is called the diffusion-matrix-rule model. This paper gives an example to show how to use the suggested model to estimate a nonlinear function with a given sample with a noise. The results from a series of simulation experiments show that, for estimating a nonlinear function, when the noise of the given sample is not low, the diffusion-matrix-rule model is better than the multi-layer back-propagation networks. In any cases, the spent time of the new model used to estimate e nonlinear function is very short. In the best case, the error of the diffusion-matrix-rule model is just one-tenth of the neural network model.