Gradual rules and extended functional dependencies-a discussion

This paper discusses the interest of gradual rules for the representation of extended functional dependencies, in classical relational databases, in the particular case of continuous domains for some attributes. First, it shows that classical rules can represent imprecise functional dependencies, but that they are not adapted to the case of a large amount of noisy or uncertain data. Then, the case of pure gradual rules is considered, in order to relax the constraints imposed by the classical rules. In this case, it is shown that a functional dependency can be more restrictive than the imprecise one it is supposed to relax. Thus, the use of special kinds of partitions, or the use of fuzzy gradual rules are studied.