Mining effective association rules using support-conviction framework

Abstract

Discovering association rules is one of the most important tasks in data mining. Most of the research has been done on association rule mining by using the support-confidence framework. In this thesis, we point out some drawbacks of the support-confidence framework for mining association rules. In order to avoid the limitations in the rule selection criterion, we replace confidence by the conviction, which is a more reliable measure of implication rules. We have generated the test data synthetically by the Hierarchical Synthetic Data Generator, which appropriately models the customer behaviour in the retailing environment. Experimental Results show that there is higher correlation between the antecedent and consequent of the rules produced by the supportconviction framework compared with the rules produced by support-confidence framework. Although support-conviction framework mines the effective associations but the association rules generated are large in numbers that are difficult to deal with. To overcome this problem, we propose an association rule pruning algorithm, which produces non-redundant and significant rules. Results obtained with synthetic data show that the proposed approach for mining association rules is quite effective and generates meaningful associations among the sets of data items.