Adaptive fuzzy classifier with a fixed number of fuzzy rules

The adaptive fuzzy system has many advantages over the neural network. It can be used to design a fuzzy classifier. However, there is the “Dimension Calamity” with the number of inputs increasing in the adaptive fuzzy system. In this paper, an adaptive fuzzy classifier with a fixed number of fuzzy rules is proposed. This classifier is combined by several fuzzy reasoning machines so that one fuzzy reasoning machine recognizes only one class. Every fuzzy reasoning machine includes two “If-Then” fuzzy rules. The total fuzzy rules number of the classifier is confirmed by the number of classes of the patterns being classified. The classifier uses the “Error Back-Propagation Training” arithmetic as the learning arithmetic. Compared with the BP neural network classifier, the new classifier and BP neural network classifier are both tested by the famous iris dataset and Ripley's synthetic dataset. It is proved that the new classifier has a good classification ability and learning ability even if the data have been polluted.