On pruning and tuning rules for associative classifiers

Abstract

The integration of supervised classification and association rules for building classification models is not new. One major advantage is that models are human readable and can be edited. However, it is common knowledge that association rule mining typically yields a sheer number of rules defeating the purpose of a human readable model. Pruning unnecessary rules without jeopardizing the classification accuracy is paramount but very challenging. In this paper we study strategies for classification rule pruning in the case of associative classifiers. © Springer-Verlag Berlin Heidelberg 2005.