Leveraging Time-Series Foundation Models for Financial Crime Detection
In briefUsing time-series foundation models to detect financial crime
SummaryThis paper proposes TCEE, a framework that summarizes each account's transaction history with a frozen time-series foundation model and adds the relevant temporal context to each transaction's representation. These representations can be used for edge classification or as edge features in a graph neural network. Across six AMLWorld datasets, TCEE improves PR-AUC by 10.8% on average over the strongest baseline on five datasets and trains 6.0 times faster end-to-end.