High confidence association mining without support pruning

Abstract

High confidence associations are of utter importance in knowledge discovery in various domains and possibly exist even at lower threshold support. Established methods for generating such rules depend on mining itemsets that are frequent with respect to a pre-defined cut-off value of support, called support threshold. Such a framework, however, discards all itemsets below the threshold and thus, existence of confident rules below the cut-off is out of its purview. But, infrequent itemsets can potentially generate confident rules. In the present work we introduce a concept of cohesion among items and obtain a methodology for mining high confidence association rules from itemsets irrespective of their support. Experiments with real and synthetic datasets corroborate the concept of cohesion. © Springer-Verlag Berlin Heidelberg 2007.