🛡️ TechFiesta Fraud Guard

AI-Powered Real-Time Fraud Detection System

A cascaded machine learning system with 30 expert rules, Bayesian Neural Network, and XGBoost/LightGBM/CatBoost ensemble achieving 83% recall at 52% precision.

📖 Documentation • 🚀 Quick Start • 📡 API • 🧠 ML Pipeline

---

🎬 Demo Video

---

🚨 The Problem

Digital payment fraud is a $50+ billion crisis in 2025, with losses projected to exceed $343 billion by 2027. Traditional fraud detection systems face a critical flaw:

Challenge	Traditional Systems	Our Solution
False Positive Rate	90-98% of alerts are false	48% false positive rate
Explainability	ML black boxes	30 interpretable rules
Adaptability	Rigid rule thresholds	Dynamic cascaded blending
Precision	2-10%	52%

Traditional AML systems waste compliance teams' time—up to 98% of alerts are non-fraudulent.

Our cascaded hybrid approach combines the best of both worlds: explainable rules for instant decisions, ML for pattern recognition, and adaptive blending based on risk level.

📖 Full Problem Statement & Solution →

---

⚡ Highlights

Metric	Value	Description
Recall	83.02%	Catches 83% of all fraud
Precision	52.52%	Half of flagged transactions are true fraud
F2-Score	74.38%	Recall-weighted performance metric
AUC-ROC	89.00%	Excellent discrimination ability

---

🎯 How It Works

Transaction → Rule Engine → BNN Risk Check → ML Ensemble → Decision
                  ↓                              ↓
            Hard Block (≥85%)           Cascaded Blending

1. 30 Expert Rules instantly flag high-risk patterns (5.7% hard-blocked)
2. Bayesian Neural Network identifies gray-zone transactions
3. 3-Model Ensemble (LightGBM 68%, XGBoost 24%, CatBoost 8%) makes final prediction
4. Cascaded Blending adjusts ML/Rules weight based on risk level

📖 Full Architecture →

---

🚀 Quick Start

Backend

cd Backend
pip install -r requirements.txt
python main.py  # Runs on http://localhost:8000

Frontend

cd Frontend
npm install
npm run dev    # Runs on http://localhost:5173

ML Training (Optional)

cd ML_Model_Training_Files
pip install -r requirements.txt
jupyter notebook  # Open the training pipeline

📖 Detailed Installation →

---

📁 Project Structure

├── Backend/           # FastAPI REST API
│   ├── main.py        # /predict endpoint
│   ├── ml.py          # FraudInferenceSystem
│   └── models/        # Trained ML models (.joblib)
│
├── Frontend/          # React 19 + Vite Dashboard
│   └── src/pages/     # 8 dashboard modules
│
├── ML_Model_Training_Files/  # Training notebooks
│   └── TechFiesta_2026_...Training_Pipeline.ipynb
│
├── images/            # 📸 Dashboard Screenshots
│
└── docs/              # 📖 Detailed Documentation
    ├── problem-statement.md # Problem & solution context
    ├── ml-pipeline.md       # ML architecture
    ├── rules-engine.md      # 30 fraud rules
    ├── api-reference.md     # API docs
    └── installation.md      # Setup guide

---

📚 Documentation

Document	Description
Problem Statement	Why this matters & our approach
ML Pipeline	Complete 8-step training architecture
Rules Engine	30 expert rules across 9 tiers
API Reference	/predict endpoint documentation
Installation	Full setup instructions

---

🖥️ Dashboard Pages

Page	Route	Description
Command Dashboard	/command-dashboard	Real-time monitoring
Detection Analytics	/detection-analytics	Fraud trends
Alert Management	/alert-management-center	Alert queue
Risk Scoring	/risk-scoring-engine	Risk tools
Fraud Form	/fraud-detection-form	Manual testing

📸 View all screenshots →

---

📄 License

MIT License - see LICENSE for details.

---

Built with ❤️ for TechFiesta 2026