Using rule order difference criterion to decide whether to update class association rules

Abstract

Associative classification is a well-known data classification technique. With the increasing amounts of data, maintenance of classification accuracy becomes increasingly difficult. Updating the class association rules requires large amounts of processing time, and the prediction accuracy is often not increased that much. This paper proposes an Incremental Associative Classification Framework (IACF) for determining when an associative classifier needs to be updated based on new training data. IACF uses the rule Order Difference (OD) criterion to decide whether to update the class association rules. The idea is to see how much the rank order of the associative classification rules change (based on either support or confidence) and if the change is above a predetermined threshold, then the association rules are updated. Experimental results show that IACF yields better accuracy and less computational time compared to frameworks with class association rule update and without class association rule update, for both balanced and imbalanced datasets. © 2010 Springer-Verlag Berlin Heidelberg.