Fraud Detection System

Designed and implemented an anomaly detection model utilizing Generative Adversarial Networks (AnomalyGAN) to identify fraudulent transactions with a precision rate of 95%. This system successfully reduced false positives by 20%.

The model was trained exclusively on legitimate transaction data, relying on reconstruction error for anomaly detection.

Integrated SHAP for model interpretability, ensuring transparency and trust in decision-making. Performance metrics such as AUC-ROC and F1-Score validated the model’s effectiveness, and the system was deployed using Flask for real-time fraud detection.