Association reducts: A framework for mining multi-attribute dependencies

We introduce the notion of an association reduct. It is an analogy to association rules at the level of global dependencies between the sets of attributes. Association reducts represent important complex relations, beyond usually considered "single attribute - single attribute" similarities. They can also express approximate dependencies in terms of, for instance, the information-theoretic measures. Finally, association reducts can be extracted from data using algorithms adapted from the domain of association rules and the theory of rough sets.