A review of data-driven approaches to improving transaction monitoring systems for enhanced detection of money laundering and terrorism financing in U.S. financial institutions

This paper is a narrative review of data science methods of improving transaction monitoring systems to combat anti-money laundering and counter-terrorism financing in the U.S. financial institutions. It takes a critical look at the shift of rule-based systems to complex data-driven models based on machine learning, network analysis, and real-time analytics, and analyzes the effect of these models on detection accuracy, false-positive reduction, and operational efficiency. The study conceptualizes the transaction monitoring process as a socio-technical system that is influenced by the interaction of analytical models, data infrastructures, governance structures and human decision-making. The review reveals the main performance improvements and ongoing issues, such as the lack of data quality, integrity, interpretability of the model, and high implementation expenses. It also discusses the implications of governance and regulation, and the increasing significance of explainability and accountability as well as strong data management practices in compliance. The changing nature of the human analysts is also emphasized, especially in the interpretation of model outputs and assisting in making decisions within hybrid systems. The results indicate that data-driven monitoring can be effective, but not universal, depending on institutional capacity, governance structure, and regulatory alignment. The paper also cites significant theoretical and empirical gaps, such as the lack of interdisciplinary integration, and absence of longitudinal and real-world evidence, and urges more holistic and mixed-method study designs.