Optimization of the AUC Criterion for Rule Subset Selection

The area under the ROC curve (AUC) is considered a relevant criterion to deal with imbalanced data, misclassification costs and noisy data. Based on this preference, we present an algorithm for rule subset selection. The algorithm builds a Pareto Front using the Sensitivity and Specificity criteria selecting rules from a large set of rules. An empirical study is carried out to verify the influence of the A priori Parameter in Pareto Front Elite Algorithm. We compare our results with other rule induction algorithms and the results show that the new algorithm obtains a set of rules with greater values of the AUC.