Mining both positive and negative association rules from frequent and infrequent itemsets

Abstract

A lot of new problems may occur when we simultaneously study positive and negative association rules (PNARs), i.e., the forms A⇒B, A⇒¬, ¬A⇒B and ¬A⇒¬B. These problems include how to discover infrequent itemsets, how to generate PNARs correctly, how to solve the problem caused by a single minimum support and so on. Infrequent itemsets become very important because there are many valued negative association rules (NARs) in them. In our previous work, a MLMS model was proposed to discover simultaneously both frequent and infrequent itemsets by using multiple level minimum supports (MLMS) model. In this paper, a new measure VARCC which combines correlation coefficient and minimum confidence is proposed and a corresponding algorithm PNAR_MLMS is also proposed to generate PNARs correctly from the frequent and infrequent itemsets discovered by the MLMS model. The experimental results show that the measure and the algorithm are effective. © Springer-Verlag Berlin Heidelberg 2007.