Target-shaped possibilistic clustering applied to local-model identification

In this paper, application of possibilistic clustering techniques to identification of local linear models will be discussed. In particular, a generalisation of some possibilistic algorithms in the bibliography is obtained. With the presented procedures, a trade-off between an “expected shape” of the membership functions and model fit can be stated. Possibilistic clustering may allow for better detection of undermodelling and overmodelling than basic techniques based on fuzzy partitions.