Multi-Level Association Rule Mining: An Object-Oriented Approach based on Dynamic Hierarchies

  • Author(s) / Creator(s)
  • Technical report TR96-15. Previous studies in data mining have yielded efficient algorithms for discovering association rules. But it is well-known problem that the two controlling measures of support and confidence, when used as the sole definition of relevant association rules, are too inclusive --- interesting rules are included with many uninteresting cases. A typical approach to this problem is to augment the thresholds with domain-specific knowledge, in the form of one or more business-goal-driven classification taxonomies. Most existing proposals use the relational approach to organize and maintain the required multi-level concept hierarchies and their domain-dependent components. We present the motivation for a necessary extension to existing uses of such multi-level hierarchies, and describe an implementation that is better suited than the typical relational paradigm for focusing the search and regulating the mining of association rules both at multi-levels within one concept hierarchy, and across multiple concept hierarchies. Our hierarchy design supports an adaptive encoding scheme for focusing the mining on semantically deeper and more informative knowledge; in essence, it dynamically generates and adjusts concept hierarchies. We demonstrate that the application of an object-oriented implementation of such a design not only provides the advantage of a flexible combination of multiple multi-level concept hierarchies for focusing the data mining task, but also provides smooth integration of multi-level concept hierarchies with legacy relational databases. In addition, by using the adaptive encoding scheme, efficient algorithms developed for discovering interesting association rules can be integrated into the our framework with no or little extra cost. | TRID-ID TR96-15

  • Date created
  • Subjects / Keywords
  • Type of Item
  • DOI
  • License
    Attribution 3.0 International