The client needed to build an ML-based solution to identify, prevent, and mitigate the risks associated with money mule activities by analyzing customer profiles and transaction patterns.

Developed and implemented a supervised ML model for money mule detection.

1. Feature Generation: Generated holistic features associated with customer profiles, including variables based on declared KYC, behavioral attributes, and application-related information (device ID, phone numbers, etc.).

2. Model Development: Utilized supervised ML algorithms, with XGBoost as the final solution, opposed to challenger models (ANN, Decision Jungles, Bayesian) trained on labeled data. The model achieved an 88% accuracy and a 92% precision rate.

3. Pattern Generation: Predicted suspicious customers using trained algorithms to generate networks for pattern visualization. Transaction patterns of mule customers were visualized to aid investigators.

4. Model Implementation: Implemented the architectural structure using cloud-based technology, automated end-to-end, and packaged in a toolkit for easy execution by the client's IT team.

Improved detection of money mule activities and saved $22M annually.

- The proactive measures protected customers and maintained the integrity of the financial system, resulting in significant annual savings.

- The model's high accuracy and precision rates enhanced the bank's ability to detect and prevent money mule activities.