The bank needed support to comply with BSA/AML guidelines, including customer/account segmentation, peer group refinement, and scenario threshold tuning.

Developed segmentation and peer group models using advanced methodologies.

1. Data Examination: Analyzed 12 months of data, including L1, L2, and SAR alerts.

2. Segmentation Model: Developed segmentation models using K-means clustering, further refining individual segments into peer groups as needed.

3. Peer Group Analysis: Designed methodologies for peer group analysis, including data wrangling, attribute selection, static time window selection, and synchronizing active and inactive accounts.

4. Model Development: Used the Mahalanobis method and compared performance using standard deviation and Manhattan distance measures. Conducted sensitivity analysis with respect to time window and peer group granularity.

5. Performance Measurement: Measured model performance using goodness of fit, ROC, average performance curve, and misclassification rate. All analyses were performed in R.

Improved AML transaction monitoring through effective segmentation.

The segmentation and peer group analysis allowed the bank to monitor customer activities more effectively, refine scenario thresholds, and enhance compliance with BSA/AML guidelines. The advanced models and methodologies improved the accuracy and effectiveness of their transaction monitoring efforts.