Fuzzy Models of Evolvable Granularity
Witold Pedrycz · 2010
Considering the inherent granularity present in fuzzy modeling, the objective of this chapter is to endow fuzzy models with an important feature of evolvable granularity—granularity whose level mirrors the varying dynamics and perception of the data/system. Depending on the architecture of the fuzzy models, the changes in granularity can be directly translated into the number of rules, local models, number of fuzzy neurons, and so on. The authors revisit and redevelop Fuzzy C-Means (FCM) so that the generic algorithm could be efficiently used in the framework of dynamic data analysis. The chapter concentrates on the fundamental design aspects, namely (1) splitting and merging criteria, (2) assessment of quality of clusters that could be directly used when controlling the dynamics of the clusters and deciding on the change of the number of clusters themselves, and (3) optimization capabilities of fuzzy clustering (FCM), which could be exploited upfront when running the algorithm. Controlled Vocabulary Terms fuzzy neural nets; merging; optimisation; workstation clusters