A kernel method for fuzzy systems modeling and approximate reasoning

Fuzzy systems modeling has been an active research topic for almost twenty years. In general, two approaches have been proposed in the literature: 1) translate experts' prior knowledge into fuzzy rules; 2) learn a set of fuzzy rules from given training data. The first approach applies to the case that prior knowledge can be easily obtained and training data are not sufficient. However, in many applications, the amount of training data is usually large. In that case, the second approach may provide more objective rules than the first approach. In this paper, we show that a class of fuzzy systems is in essence kernel machines. Therefore, the support vector machine (SVM) method can be used to construct fuzzy systems. This method has been tested on real weather forecast data. Experimental results demonstrate the effectiveness of the method.