Anomaly Detection in Financial Transactions Via Graph-Based Feature Aggregations

Anomaly detection in the financial domain aims to detect abnormal transactions such as fraudulent transactions that can lead to loss of revenues to financial institutions. Existing solutions utilize solely transaction attributes as feature representations without the consideration of direct/indirect interactions between users and transactions, leading to limited accuracy. We formulate anomaly detection in financial transactions as the problem of edge classification in an edge-attributed multigraph, where each transaction is regarded as an edge, and each user is represented by a node. Then, we propose an effective solution DoubleFA, which contains two novel schemes: proximal feature aggregation and anomaly feature aggregation. The former is to aggregate features from neighborhoods into edges based on top-k Personalized PageRank (PPR). In anomaly feature aggregation, we employ a predict-and-aggregate strategy to accurately preserve anomaly information, thereby alleviating the over-smoothing issue incurred by proximal feature aggregation. Our experiments comparing DoubleFA against 10 baselines on real transaction datasets from PayPal demonstrate that DoubleFA consistently outperforms all baselines in terms of anomaly detection accuracy. In particular, on the full PayPal dataset with 160 million users and 470 million transactions, our method achieves a significant improvement of at least 23% in F1 score compared to the best competitors.