||Data-driven fuzzy modeling is concerned with the induction of computational models from data. When creating fuzzy models from data, the problem arises how to choose the underlying fuzzy sets. On the one hand, the fuzzy sets should have a semantic meaning to ease interpretation, but on the other hand, their shape has a large influence on the quality of the resulting fuzzy model.In this paper, we will present an algorithm to derive fuzzy partitions from data. We will then illustrate the influence of the number and shape of fuzzy sets on the quality and the complexity of the resulting models. We show that by using ordering-based predicates, the problem of choosing the optimal number of fuzzy sets can be overcome.Finally, we will give an outlook on post optimization of fuzzy rule bases.