Work

Machine Learning and Network Analysis for Cross-Border Suspicious Transaction Detection: Evidence From the FinCEN Files

Year: 2026

Type: article

Abstract: Traditional rule-based anti-money-laundering (AML) systems raise far more alerts than humans can plausibly investigate. In the aftermath of the International Consortium of Investigative Journalists' FinCEN Files revelations, compliance teams and regulators have begun to ask whether a data-driven approach might allocate scarce attention more effectively. This study presents a holistic framework that combines descriptive statistics, time-series modelling, geospatial analysis, network science and machine-learning classification to analyse 4,507 aggregated transaction groups (more than 200,000 individual transaction records) from the FinCEN Files dataset. After cleaning and integrating the data, we explore the heavy-tailed distribution of suspicious amounts, identify temporal spikes linked to economic crises, map cross-border corridors between secrecy jurisdictions, quantify centrality in the bank-to-bank network, fit an ARIMA model to monthly volumes, and construct a risk-scoring system. Logistic regression, Random Forest and XGBoost models are benchmarked against unsupervised isolation and one-class support vector methods. Ensemble models achieve area-under-the-ROC-curve scores around 0.97-0.98, substantially outperforming rule-based or unsupervised baselines. Our results demonstrate that network metrics, particularly degree and betweenness centrality-coupled with high-risk corridor indicators are among the most important predictors of suspicious behaviour. We conclude by situating these findings within the emerging literature on graph neural networks and proposing avenues for continual learning, beneficial ownership integration and cross-institutional transferability. Overall, the work illustrates that a judicious blend of machine learning and network analysis can reduce false positives and support risk-based AML supervision.

Source: IEEE Access

DOI: 10.1109/access.2026.3685103

Authors: Oghenefejiro M. Ejime, Abiola Kosemani, Shedrack Isioma Egede Onyeorueka

Institution: University of Salford

Language: English

Topic: Crime, Illicit Activities, and Governance

Subfield: Sociology and Political Science

Field: Social Sciences

Domain: Social Sciences

SDG: Peace, justice, and strong institutions

Open Access status: gold

Estimated APC paid: $2,075