SilIF: Silhouette-Augmented Isolation Forest for Unsupervised Transaction Fraud Detection

In three linesSilIF augments Isolation Forest for fraud detection by adding a silhouette-based scoring layer computed from tree path lengths. On IEEE-CIS benchmark (~590K transactions, 3.5% fraud), SilIF achieves +0.0080 AUC-PR improvement over plain IF (p=0.046). No gains on Sparkov dataset; paper characterizes when the augmentation helps.

Summary generated by Claude — human-verified