Class association rule mining with correlation measures using genetic network programming
Eloy Gonzales, Shingo Mabu, Karla Taboada, Kaoru Shimada, Kotaro Hirasawa · 2009 ICCAS-SICE · 2009
Association rule mining is one of the tasks of data mining and it has been extensively studied recently. As a consequence, several methods for extracting association rules have been developed during the last years. Most of them use the support and confidence framework to extract the association rules. Researches are able to extract strong rules using this framework. However these measures are not good enough to solve the quality problems of the rules. A new data mining method using Genetic Network Programming (GNP) has also been developed recently which uses the χ f2 (chi-squared) as a correlation measure and its effectiveness has been shown for different datasets [1] [2]. To enhance the correlation degree and comprehensibility of association rule, several correlation measures including lift, χ f2 , all-confidence and cosine are studied in this paper when they are incorporated in the conventional GNP based mining algorithm. A comparison between the correlation measures is made in the simulations when they are incorporated separately into the GNP based mining method. Finally, the association rules extracted using different correlation measures are applied to the classification problems and the prediction accuracies of them are evaluated.