A Comparison of Methods for Rule Subset Selection Applied to Associative Classification

This paper presents Garss, a new algorithm for rule subset selection based on genetic algorithms, which uses the area under the ROC curve – AUC – as fitness function. Garss is a post-processing method that can be applied to any rule learning algorithm. In this work, Garss is analysed in the context of associative classification, where an association rule algorithm generates a set rules to be used as a classifier. An experimental evaluation was performed in order to analyse the behaviour of the proposed method. Results are compared with Roccer, a recently proposed algorithm for rule subset selection based on ROC analysis.