Extraction of typical and exceptional fuzzy rules from data including qualitative and quantitative attributes

We propose a method for extracting two kind rules, typical and exceptional fuzzy rules, from data including quantitative attributes and qualitative attributes such as countries and sex. First, we review representing qualitative data by enumerated fuzzy sets in fuzzy neural networks and tune then by backpropagation with forgetting facility. We apply it to typical and exceptional learning methods. To extract typical rules, we generate, tune and prune a fuzzy neural network with the training data and select correct data from the training data with the trained fuzzy neural network, and we again generate, tune and prune a new fuzzy neural network with only the correct data. Its process is repeated. We extract exceptional rules from the incorrect data. The proposed method is applied to sample data for estimating human weight and real data for evaluating kitchens.