📅 Date: 27.07.2025
👤 Author: Juniyad Tamboli, Kapurhol, Pune

This project delivers a comprehensive, end-to-end machine learning pipeline to detect fraudulent transactions in financial data, leveraging the power of the XGBoost classifier. It covers everything from preprocessing to model deployment, with top-notch accuracy and AUC, ensuring robust fraud detection for real-world applications.

• 🔍 Accurate Fraud Prediction: Build a robust ML model for identifying fraudulent transactions.
• ⚖️ Imbalance Handling: Use SMOTE to boost recall for fraud cases (minority class).
• 🛠️ Model Optimization: Tune XGBoost for optimal ROC-AUC performance.
• 💡 Interpretability: Highlight and explain the most influential fraud features.
• 🚀 End-to-End Pipeline: Deliver a ready-to-deploy model and web application.

• 📄 File: fraud_dataset.csv
• 🔢 Records: ~10,000 transactions
• 🧩 Features:
  • Categorical (encoded):
    • merchant_category
    • customer_location
    • device_type
  • Numerical (scaled):
    • previous_transactions
    • customer_age
    • amount
  • 🚫 Dropped:
    • transaction_id, customer_id, timestamp

1. 🔠 Label Encoding: All categorical features
2. 📏 Feature Scaling: Normalize numerical values (StandardScaler)
3. ⚖️ Class Balancing: SMOTE for fraud rate <15%

• Algorithm: XGBoost Classifier
• Hyperparameter Tuning: GridSearchCV on:
  • n_estimators: [150,max_depth: [4, - learning_rate`: [0.03, 0.07, 0.1]
• Validation: Stratified 3-fold Cross-Validation
• Data Split: 80/20 stratified (class preservation)

precision    recall  f1-score   support

0   0.9751     0.9985    0.9867   1963
1   0.9984     0.9745    0.9863   1962

Accuracy:                     0.9865   3925
Macro avg:                    0.9868   3925
Weighted avg:                 0.9868   3925

• 🏅 ROC-AUC Score:
  0.9971659708797549

{
  'learning_rate': 0.1,
  'max_depth': 8,
  'n_estimators': 200,
  'scale_pos_weight': 0.99
}

A user-friendly Flask-based web interface enables real-time fraud detection:

• 🔑 Required Features:
  • merchant_category
  • previous_transactions
  • customer_location
  • device_type
  • customer_age
  • amount
• ⚡ Instant Results:
  • Predicts Fraudulent or Not Fraudulent 🟢🔴 with a probability score!
• 🖥️ File Structure:
  • app.py – Flask server
  • index.html – User interface
  • Fraud_Detection_System.joblib – Trained model
  • scaler.pkl – StandardScaler
  • fraud_dataset.csv – Dataset reference

1. 📦 Requirements

   • Python 3.7+
   • Libraries: flask, numpy, scikit-learn, xgboost, joblib

2. ▶️ Run the App

   python app.py

   Open your browser to http://127.0.0.1:5000 🚀

3. 🗃️ Deployment

   • Place Fraud_Detection_System.joblib & scaler.pkl in the working directory

🎉 Your fraud detection system is ready for use!
Built with XGBoost, the model demonstrates exceptional accuracy and transparency, fit for real-world banking and finance needs.
For questions or collaboration, reach out to the author.

👨💻 Author:
Juniyad Tamboli, Pune

Feel free to share, star, or reach out for partnerships! 🚀