Abstract
In this paper, an analytical framework for association rule mining based on the Dempster-Shafer (DS) evidential reasoning is proposed. The method we propose associates itemsets in a database with basic probability assignments (bpas) encountered in DS theory to numerically quantify the complex interrelationships that exist among the itemsets, thus incorporating the subjective human reasoning that may otherwise be unaccounted for. In order to recast an association within this framework, measures of support and confidence in association rule mining derived via certain conditional notions are used. These measures utilize the associated subjective knowledge of the itemsets in order to discover the interesting patterns as opposed to a simple measure of frequency of occurrence of itemsets. The manner in which the frequency of occurrence is used in the existing methods also fail to capture the associations generated by the multiplicity of an item. However the method we propose uses the subjective assignment of a bpa in order to address this issue. The association rules thus formed capture the qualitative nature of the relationships among itemsets in the database which is not sufficiently well captured in the traditional data mining analysis methods.
