Fuzzy logic-based clustering techniques are widely used in situations where statistical assumptions are not valid. Whether in estimating cluster centers for model identification purposes or in determining clusters the existing techniques are essentially based upon the choice of some potential functions. As in any design problems of this kind, the choice of such a function has to be justified on a theoretical basis. In this work, we set up a decision frame work and show that optimal potential functions are the ones which are used in current techniques.
Technical Report: UTEP-CS-98-2
Published by The Chinese University of Hong Kong, Department of Mechanical and Automation Engineering, as Technical Report CUHK-MAE-98-001, January 1998.