Improving Association Rule Mining for Infrequent Items Using Direct Importance Estimation

Abstract

This paper proposes a method to find association rules for infrequent items. Despite the long history of association rule mining, infrequent items have been usually ignored. Recently, owing to the online nature of most systems, tackling infrequent items has become increasingly important to find emerging information. The proposed method not only has a sound theoretical background but is an exact solution of error minimization. Although highly similar to the standard method, Apriori, the solution uses a different formula than Apriori. Moreover, it consistently outperforms Apriori.