MetaFraud: A Meta-Learning Framework for Detecting Financial Fraud

This appendix reports the results for the baseline and yearly/quarterly context-based classifiers when using the 1:10 regulator cost setting. Since the AUC values are computed across different cost settings (and are therefore the same for the investor and regulator situations), we report only the legitimate/fraud recall rates. Overall AUC values as well as results for the investor cost setting (1:20) can be found in the subsection “Comparing Context-Based Classifiers Against Baseline Classifiers” of the main paper.