Rule Extraction from Support Vector Machines: Measuring the Explanation Capability Using the Area under the ROC Curve

Recently, the area of rule extraction from support vector machines (SVMs) has been explored. One important indication of the success of a rule extraction method is the performance of extracted rules as compared to the original SVM. In this paper, we describe the use of the area under the receiver operating characteristics (ROC) curve (AUC) to assess the quality of rules extracted from an SVM. In particular, we directly compare AUC to the more commonly used measures of accuracy and fidelity and show that AUC is both a more reliable and meaningful measure to use