Mining Association Rules with Uncertain Item Relationships

In order to enhance the precision in association rule mining, an extension is proposed to capture the uncertain item relationships in the data sets in this paper. Two sources of uncertainty are considered: the degree of individual item importance (multiplicity) and the degree of association among the items (inter-relationship). In many real-world applications, especially in a distributed environment, the data sets are generated and collected from different sources. Thus the inter-relationships among the items can vary and result in uncertain item relationships. The Dempster-Shafer (DS) evidential reasoning theory is applied to generate the association rules with the proposed support and confidence measures under uncertainty. These measures are defined under Shannon-like measure of total uncertainty. A numerical example based on market basket analysis is given with a comparison between our approach and the original association rule mining method.