Abstract
This paper proposes a new fuzzy clustering technique to detect a number of local linear varieties based on a given data set. The main proposal is the introduction of an objective function that aims at obtaining an appropriate dimensional linear variety corresponding to the data distribution within each cluster. Then, the paper proposes a new type of fuzzy models expressed by elliptic type membership functions. One can develop this kind of models based on the clustering results only, without considering the model structure. An interactive simulation using the developed model helps us understand the data distribution in the multi-dimensional space and the relationships between variables
