Type-Aware Evaluation Improves Anomaly Detection in Financial Journal Entries

Standard ROC-AUC metrics fail to account for review budgets in financial anomaly detection. This paper introduces Fair-Share Type Recall (FSR), which reshapes detector rankings by capping credit based on budget shares, favoring models like kNN over autoencoders in practical ledger testing.