Money laundering remains a major challenge for financial institutions and regulators, requiring scalable and intelligent monitoring solutions. Traditional detection methods often rely on static rules and manual investigations, which can be inefficient when dealing with large transaction volumes and evolving fraud patterns.
This project combines data engineering, machine learning, and interactive visualization to create an AI-powered AML analysis platform. The system processes transaction data, transforms customer activity into behavioral indicators, and applies anomaly detection models to identify potentially suspicious behavior.
The system generates an output file containing customer risk scores alongside explainable insights that highlight the main behaviors contributing to each classification. This approach helps analysts prioritize investigations more efficiently while maintaining transparency throughout the risk assessment process.
The platform transforms raw transaction data into explainable AML risk scores through a multi-step analysis pipeline. Transaction data is first processed and aggregated into customer behavioral profiles. Multiple analytical models then evaluate customer activity from different perspectives, identifying unusual patterns and behavioral anomalies.
The outputs from these models are combined into a final risk score, along with explanations highlighting the main factors that contributed to the classification.
Instead of analyzing transactions individually, the model transforms customer activity into behavioral indicators that summarize financial behavior over time.
The analyzed features are organized into different behavioral dimensions, including:
Transaction Intensity: total number and monetary volume of sent and received transactions.
Transaction Size Patterns: typical transaction values, including median transaction amounts.
Transaction Variability: variability and maximum values of financial operations.
Fund Flow Patterns: balance between incoming and outgoing transactions and funds.
Short-Term Activity: recent transaction behavior over 7-day and 30-day periods.
Activity Acceleration: sudden increases or decreases in transaction volume and frequency over time.
These indicators allow the system to identify customers whose behavior significantly differs from expected financial activity patterns.
The final AML risk score is generated through a multi-layer anomaly detection approach, where each model captures a different dimension of suspicious behavior.
The methodology combines:
Global anomaly detection: identifies customers with behavior that is rare compared to the entire population.
Behavioral clustering: detects customers that do not clearly belong to common behavioral groups.
Peer-group deviation analysis: measures how much a customer deviates from others with similar profiles.
The outputs of these layers are normalized and combined into a final score that prioritizes customers for further investigation.
The objective of the system is not to directly classify fraud, but to identify behaviors that are unusual enough to deserve additional review by compliance analysts.
The platform generates an output file containing the AML risk assessment for each analyzed customer. Each record includes a final risk score and explainable insights describing the main behavioral patterns that influenced the classification.
Clustering
[1] Chandola, Varun, Arindam Banerjee, and Vipin Kumar. "Anomaly detection: A survey." ACM computing surveys (CSUR) 41.3 (2009): 1-58. Access on: https://dl.acm.org/doi/abs/10.1145/1541880.1541882
[2] Knox, Edwin M., and Raymond T. Ng. "Algorithms for mining distancebased outliers in large datasets." Proceedings of the international conference on very large data bases. Citeseer, 1998. Access on: https://www.researchgate.net/publication/2346156_Algorithms_for_Mining_Distance-Based_Outliers_in_Large_Datasets
[3] Jensen, Rasmus Ingemann Tuffveson, and Alexandros Iosifidis. "Fighting money laundering with statistics and machine learning." Ieee Access 11 (2023): 8889-8903. Access on: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10025710
Dimension reduction
[4] Bakry, Ahmed N., et al. "Combating financial crimes with unsupervised learning techniques: clustering and dimensionality reduction for anti-money laundering." arXiv preprint arXiv:2403.00777 (2024). Access on: https://arxiv.org/pdf/2403.00777
System architecture
[5] Sizan, Mir Mohtasam Hossain. "Machine learning-based unsupervised ensemble approach for detecting new money laundering typologies in transaction graphs." International Journal of Applied Mathematics 38.2s (2025): 351-374.