💳 Real-Time Credit Card Fraud Detection System

📌 Overview

A scalable, production-ready machine learning system for detecting fraudulent credit card transactions in real time.
The project focuses on handling extreme class imbalance, optimizing precision-recall tradeoffs, and providing model explainability for decision transparency.

---

🎯 Problem Statement

Credit card fraud detection presents key challenges:

• Highly imbalanced datasets (fraud cases are rare)
• Need for low-latency predictions in real-time systems
• Requirement for interpretable models in financial applications

This system addresses these challenges using a robust ML pipeline and API-based deployment.

---

🧠 Solution Architecture

Raw Data → Preprocessing → SMOTE → Model Training (XGBoost)
         → Hyperparameter Tuning (Optuna)
         → Evaluation (Precision-Recall Optimization)
         → Explainability (SHAP)
         → Deployment (FastAPI)

---

📊 Performance

Metric	Score
Precision	93%
Recall	85%

• Optimized using precision-recall tradeoff
• Focused on minimizing false positives while maintaining strong recall

---

⚙️ Key Features

• Imbalance Handling

  • SMOTE-based oversampling for minority class

• Model Optimization

  • Hyperparameter tuning using Optuna

• Explainability

  • SHAP for feature-level prediction insights

• Real-Time Inference

  • FastAPI-based REST API for low-latency predictions

• Production-Oriented Design

  • Modular code structure
  • Model serialization using joblib

---

🛠 Tech Stack

• Languages: Python
• ML: XGBoost, Scikit-learn
• Data: Pandas, NumPy
• Imbalance Handling: imbalanced-learn (SMOTE)
• Optimization: Optuna
• Explainability: SHAP
• Backend: FastAPI
• Serving: Uvicorn

---

📂 Project Structure

.
├── api/
│   └── app.py              # FastAPI application
├── models/
│   └── model.pkl          # Trained model
├── src/                   # Core ML pipeline
├── notebooks/             # Experiments and analysis
├── requirements.txt
└── README.md

---

🚀 Getting Started

1. Clone Repository

git clone <your-repo-url>
cd fraud-detection

2. Install Dependencies

pip install -r requirements.txt

3. Run the API Server

uvicorn api.app:app --reload

---

🔌 API Usage

Endpoint

POST /predict

Sample Request

{
  "features": [0.1, -1.2, 3.4, ...]
}

Sample Response

{
  "fraud": true,
  "probability": 0.92
}

---

📈 Model Explainability

• Uses SHAP (SHapley Additive Explanations)
• Provides per-transaction feature importance
• Enables:
  • Debugging model predictions
  • Auditability for financial systems

---

🧪 Future Improvements

• Real-time streaming (Kafka / Flink)
• Model monitoring and drift detection
• Docker containerization
• Kubernetes deployment
• Feature store integration
• CI/CD pipelines

---

👨‍💻 Author

Utsav Kashyap