A ML approach to Financial Crime Detection: A Study on Transaction Monitoring Models in AM

AML frameworks are essential for protecting financial systems against illicit funds. This thesis focuses on Transaction Monitoring (TxM), the cornerstone of AML compliance, which involves the continuous surveillance of customer transactions to spot suspicious patterns indicative of money laundering or terrorist financing. First, we provide an overview of AML’s core components—Know Your Customer (KYC)/Customer Due Diligence (CDD), Name Screening, and Risk Rating—highlighting how TxM complements these preventive measures by acting as a detective control. Next, we examine modern TxM systems, contrasting traditional rule-based approaches with advanced analytics and machine learning techniques. Rule-based scenarios rely on expert-defined thresholds and red-flag patterns, while machine learning models enable anomaly detection and alert prioritization through supervised and unsupervised algorithms. Network and graph analytics uncover hidden relationships among accounts and entities, exposing laundering networks that simple rule logic would miss. A central challenge in TxM is scenario calibration: setting and tuning thresholds to balance true positives against false positives. We discuss data-driven calibration methodologies—Above-the-Line (ATL) and Below-the-Line (BTL) testing—used for initial deployment and periodic retuning of rules. Key scenario categories include high-risk geography transactions, structuring or “smurfing,” behavioral anomalies (e.g., rapid fund movements), and network-based hidden relationships. Practical examples illustrate how properly calibrated scenarios detect structuring, large high-risk wire transfers, and potential terrorist financing via charities. The thesis also reviews leading platforms such as Oracle Financial Crime and Compliance Management (FCCM) and Compliance Studio, which integrate rule engines, machine learning, and case management to automate and streamline the AML process. Best practices and regulatory expectations emphasize a risk-based approach, robust governance, and continuous improvement to address evolving typologies. Through this comprehensive exploration, the thesis demonstrates how sophisticated TxM systems, combined with rigorous calibration and governance, form a dynamic defense against financial crime while meeting regulatory requirements.