An Effective Algorithm for Mining Interesting Maximal Association Rules

Abstract

Most existing algorithms focus on Association Rules Mining (ARM) based on a traditional support-confidence framework. These algorithms produce a large number of redundant rules, the majority of which are irrelevant to the users or do not imply a correlation relationship between related itemsets. In this paper, an effective algorithm that incorporates the generation of Maximal Frequent Itemsets (MFIs) that ensures removal of redundancy and correlation analysis has been adopted as an interesting measure is suggested. The proposed algorithm integrates the support-all-confidence measures as a new constraint framework to be pushed deep during the mining process of MFIs to generate a reduced and complete set of All-Confident Correlated Maximal Frequent Itemsets (ACCMFIs) directly from large datasets. Consequently, the generated ACCMFIs are considered as a new basis for the discovery of Interesting Maximal Association Rules (IMARs). The proposed algorithm has been developed, and the experimental results demonstrate its utility and effectiveness.