A backward identification approach to fuzzy rules learning using a centroid-resonance neural network

A new method is developed to generate fuzzy rules from numerical data. This new method consists of two main steps: Step 1 divides the output spaces of the given numerical data into fuzzy clusters by unsupervised learning using a centroid-resonance neural network (CRNN). Step 2 regards the acquired degree of membership as a target signal, and uses it to identify the structure of the input spaces by a backpropagation algorithm. Consequently, one can acquire such fuzzy rules which are driven by an artificial neural-network in their premise parts and are real numbers in their consequence parts, where the numbers are considered to be fuzzy numbers representing centroids of the acquired fuzzy clusters. Therefore, the authors call this method the backward identification method.