Discovering Interesting Patterns in Numerical Data with Background Knowledge

Association rule mining (Agrawal, Imielinski & Swami, 1993) is one of the most important data mining tasks. Initially only simple conjunctions of items were allowed as patterns, but generalizations of the framework to other pattern types such as sequences, trees, graphs etc. have been developed, significantly expanding its applicability. Curiously there has been relatively little effort devoted to generalizing association rules to numerical attributes, despite the practical ubiquity of numerical data. The main approach to mining numerical data has been discretization (Srikant & Agrawal, 1996). In this approach numerical attributes are split into a number of discrete intervals, after which the data can be mined using standard techniques. Discretization however has several problems. First, discretizing the attributes leads to information loss. Second, each interval contains only a small portion of the data which can lead to statistical estimation problems. The third problem is that relationships between atAbstrAct