Association Rules for Expressing Gradual Dependencies

Data mining methods originally designed for binary attributes can generally be extended to quantitative attributes by partitioning the related numeric domains. This procedure, however, comes along with a loss of information and, hence, has several disadvantages. This paper shows that fuzzy partitions can overcome some of these disadvantages. Particularly, fuzzy partitions allow for the representation of association rules expressing a tendency, that is, a gradual dependence between attributes. This type of rule is introduced and investigated from a conceptual as well as a computational point of view. The evaluation and representation of a gradual association is based on linear regression analysis. Furthermore, a complementary type of association, expressing absolute deviations rather than tendencies, is discussed in this context.