Mining association rules on related numeric attributes

Abstract

In practical applications, some property is represented by a pair of related attributes. For example, blood pressure, temperature changes etc. The existing data mining approaches for association rules can not tackle those cases, because they treat every attribute independently. In this paper, as a special kind of correlation, we express the pair of attributes as a range-type attribute. We define a set of fuzzified relations between ranges and revise the definition of association rules. We also propose effective algorithms to evaluate the measures for ranking association rules on related numeric attributes.