Time-Series Foundation Models Enhance Financial Crime Detection
ORIGINAL / Leveraging Time-Series Foundation Models for Financial Crime Detection
The study proposes TCEE, a framework that uses a frozen time-series foundation model to summarize account transaction histories and enrich edge representations with temporal context, improving PR-AUC and training speed on multiple anti-money laundering datasets, offering a new approach for financial crime detection.
01 ABSTRACT
Financial crime detection requires identifying rare illicit activities that unfold over time within large transaction networks. Existing graph-based methods capture account relationships but often underuse temporal behavior in transaction histories. The authors introduce Temporal-Context Edge Enrichment (TCEE), which uses a frozen time-series foundation model (TSFM) to summarize account histories and enrich transaction representations, achieving an average relative PR-AUC improvement of 10.8% on five of six AMLWorld datasets and 6x faster training.
02 KEY FINDINGS
- TCEE uses a frozen TSFM to summarize account transaction histories and provide temporal context features for edge classification.
- On six AMLWorld datasets, TCEE improves PR-AUC by an average relative 10.8% on five datasets and achieves 6.0x faster end-to-end training.
- Feeding TCEE representations into a GNN further improves PR-AUC by 4.8% on four smaller datasets; on an Ethereum phishing dataset, node classification PR-AUC improves by 14.5%.
AI GENERATED SUMMARY / DISCOVERED BY ARXIV CS.LG