Extracting representative measures for the post-processing of association rules

摘要

This paper deals with finding a minimum set of interestingness measures in the stage of post-processing of association rules. These measures, called representative measures, are calculated with the help of a medoid clustering. The main interest of this approach is to deliver a reduced set of measures that is specific and adapted to each dataset tudied. The result obtained also facilitates the validation of the best rules. Furthermore, the approach is applied to a rule-based dataset of 123228 association rules with thirty-six measures. As a result of this dataset, we obtain a reduced set of sixteen representative measures. The paper also summarizes the state-of-the-art post-processing and the relative works about interestingness measures.