About

Built a real-time credit card fraud detection system using Isolation Forest anomaly detection on 284,807 transactions.
Achieved 32.7% fraud detection with 0.19% false alarms, delivering a 93.7% ROI by linking model outputs to business impact.
Performed feature importance analysis to interpret PCA-transformed features and deployed the model as a FastAPI endpoint for production-ready scoring.
Optimized for extreme class imbalance, cost-effectiveness, and scalable real-time fraud prevention.