Credit Card Fraud Detection — XGBoost + Streamlit Live Demo

An end-to-end fraud detection project that trains an XGBoost model for tabular transaction data and deploys a live Streamlit demo for scoring new transactions from a CSV upload.

Live Demo: https://credit-card-fraud-detection-model-nadine-shill.streamlit.app/
Notebook: Fraud Detection 2.ipynb

---

Overview

Credit card fraud is a classic highly imbalanced classification problem: fraudulent transactions are rare, but the impact is high. This project focuses on building a practical pipeline for:

• Data preprocessing + feature engineering
• Training an XGBoost classifier for fraud prediction
• Persisting model artifacts for reproducible inference
• Deploying an interactive Streamlit app:
  • Load a sample CSV
  • Upload your own CSV (up to 200MB)
  • Adjust fraud threshold
  • Download predictions as a results CSV

---

What’s Included

Streamlit App (Live Demo)

The demo app (app.py) allows you to:

• Load sample_input.csv (included in the repo)
• Upload a CSV with the expected features
• Generate:
  • fraud_probability (0–1)
  • fraud_prediction (0/1 based on the threshold slider)
• Download the scored results as a CSV

Model Artifacts

Saved to the artifacts/ folder:

• scaler.joblib — preprocessing scaler used at inference
• feature_columns.json — expected feature schema/order
• xgboost.json — trained XGBoost model (Booster)

---

Repo Structure

.
├── app.py
├── requirements.txt
├── sample_input.csv
├── artifacts/
│   ├── scaler.joblib
│   ├── feature_columns.json
│   └── xgboost.json
├── Fraud Detection 2.ipynb
└── README.md