EvoCMAR: A New Evolutionary Method to Directly Mine Association Rules for Classification

In this paper, we propose an evolutionary method with a three-layer structure to directly mine association rules for classification. The association rules have been demonstrated to be useful for classification, such as classification based on association rule (CBA) and classification method based on multiple association rule (CMAR), and they are found to be more accurate than some traditional methods, such as C4.5. Generally speaking, there are two phases in an associative classification method: (i) association rules mining; (ii) classification by association rules. However, the two phases are almost separated, viz, during the first phase, the mining of association rules does not focus on classification. Moreover, when building the classifier in the second phase, most of the association rues will be pruned. As a result, if we are able to directly mine the classification association rules, we can save time. Meanwhile, we can expect even better accuracy because the mining procedure itself considers the classification. In this paper, we build a novel evolutionary method, named evolutionary classification method based on multiple association rule (EvoCMAR), to tackle these problems, and the simulation results show that it performs well in both accuracy and speed. © 2010 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.