Qihan1997/fraud_detection_explainability

🛡️ FraudGuard AI Analyst Portal

This application provides a real-time Ensemble Fraud Detection interface using LightGBM, Random Forest, and a Ridge Meta-Model.

Key Features:

• —Ensemble Prediction: Blends multiple model scores for higher precision.
• —Stateful Feature Engineering: Tracks user transaction history within the session to calculate rolling Z-scores.
• —Cold-Start Handling: Uses global training priors for new customer IDs.
• —SHAP Explainability: Provides a waterfall plot to show exactly why a transaction was flagged.

How to use:

1. 1.Enter a Customer ID.
2. 2.Input transaction details like Amount and MCC.
3. 3.Click Analyze to see the probability verdict and the explanation chart.