Linguistic Rule Extraction From a Simplified RBF Neural Network

Representing the concept of numerical data by linguistic rules is often desirable. In this paper, we present a novel rule-extraction algorithm from the radial basis function (RBF) neural network classifier for representing the hidden concept of numerical data. Gaussian function is used as the basis function of the RBF network. When training the RBF neural network, we allow for large overlaps between clusters corresponding to the same class, thereby reducing the number of hidden units while improving classification accuracy. The weights connecting the hidden units with the output units are then simplified. The interval for each input in the condition part of each rule is adjusted in order to obtain high accuracy in the extracted rules. Simulations using some bench-marking data sets demonstrate that our approach leads to more accurate and compact rules compared to other methods for extracting rules from RBF neural networks.