A neuro-fuzzy approach in parts clustering

In the feature-based clustering system, conversion from part feature to crisp codes is a conventional procedure in part clustering. However, part characteristics are reduced in the procedure especially for fuzzy and interval attributes. To remedy the shortages, a self-organizing map network with fuzzy weights is proposed. By taking the linguistic representation capabilities of fuzzy theory and the clustering abilities of self-organizing map (SOM) networks, the proposed approach is able to deal with not only crisp attributes but also interval attributes as well as fuzzy attributes. Due to the fuzzy data and fuzzy weights between the input layer and output layer, a learning rule is presented. The influence of the three parameters in the network is discussed