AML Model Development | Scotiabank Case Competition
Designed an end-to-end AML detection pipeline on transaction-level data using Logistic Regression and XGBoost, achieving a 25% increase in suspicious activity capture rate compared to baseline model while balancing detection sensitivity and false positive alerts. Conducted exploratory analysis to assess data quality, transaction patterns, and high-risk behavior, informing feature engineering and model selection under severe class imbalance. Applied model interpretability techniques (SHAP) to explain model decisions, support regulatory compliance, and enhance model transparency