Anti-Money Laundering ML Modelling Approach
summary
Raiffeisen builds an unsupervised AML system that profiles customers (static + monthly dynamic features), creates peer-group comparison, and flags outliers using PCA outlier score (orthogonal and score distances combined) to rank customers into green/yellow/red buckets. Red is ~1% of customers and the project targets ~50% true-suspicious rate inside red; current work is a POC, not in production. SHAP is used for explainability and findings feed back to update rule-based alerts and supervised models.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:20Money laundering problem
- 3:26AML detection challenges
- 5:54Supervised alert classification
- 6:35Unsupervised non-alerted customers
- 10:08Normal behavior with peer groups
- 12:18PCA outlier detection
- 14:04Model effectiveness with compliance
- 15:32Q&A
related talks
23mChallenges in using machine learning for financial monitoring
46mML for Anti-Money Laundering
49mEnhanced AML fraud detection solutions with Azure Machine Learning
58mNew-type Scoring Models: Analyzing Customer Actions
01h 01mD1T2 - Applying Machine Learning to User Behavior Anomaly Analysis
27mData Science and ML with an emphasis on Banking...
27mAshrith Barthur, Security Scientist, H2o.ai at MLconf Seattle 2017
35mSHAPtivating Insights: unravelling blackbox AI models