<<<<<<< HEAD

🔐 Credit Card Fraud Detection – End-to-End Machine Learning Pipeline

An industry-grade end-to-end Machine Learning project that detects fraudulent credit card transactions using advanced ML models and serves real-time predictions via a FastAPI REST API.

This project includes:

• Data Cleaning & Feature Engineering
• Class Imbalance Handling with SMOTE
• Model Training with Random Forest, XGBoost, and SVM
• Hyperparameter Tuning
• Final Model Evaluation with ROC & Precision-Recall Curves
• Production-Ready API Deployment using FastAPI

---

🚀 Tech Stack

• Python 3.10+
• Scikit-learn
• XGBoost
• Imbalanced-learn (SMOTE)
• FastAPI
• Matplotlib / Pandas / NumPy
• Joblib

---

📂 Project Structure

fraud-detection-ml/ │ ├── data/ │ └── processed/ │ ├── clean_data.csv │ ├── train_data.csv │ └── test_data.csv │ ├── notebooks/ │ ├── 01_eda.ipynb │ ├── 02_feature_engineering.ipynb │ ├── 03_model_training.ipynb │ └── 04_hyperparameter_tuning.ipynb │ ├── models/ │ ├── tuned_random_forest.pkl │ ├── tuned_xgboost.pkl │ ├── tuned_svm.pkl │ └── feature_order.txt │ ├── app/ │ └── app.py │ ├── src/ │ └── final_evaluation.py │ ├── requirements.txt └── README.md

---

📊 Dataset

• European Credit Card Fraud Dataset (Kaggle)
• 284,807 transactions
• Only 0.17% fraud cases → highly imbalanced dataset

---

✅ Machine Learning Pipeline

1️⃣ Data Cleaning

• Duplicate removal
• Standard Scaling of numerical features
• Train-test split with stratification

2️⃣ Feature Engineering

• Correlation-based feature elimination
• Class imbalance handling using SMOTE

3️⃣ Model Training

• Random Forest
• XGBoost
• Linear SVM (for large-scale efficiency)

4️⃣ Hyperparameter Tuning

• RandomizedSearchCV for RF & XGBoost
• GridSearchCV for SVM
• Tuning optimized for maximum Recall

5️⃣ Final Evaluation

• Precision, Recall, F1-Score
• ROC-AUC
• Precision–Recall Curves
• Business Expected Loss Analysis

---

🧠 Final Model Performance

Model	ROC-AUC
Random Forest	1.000
XGBoost	1.000 ✅ (Production Model)
SVM	0.989

✅ XGBoost was selected as the production model.

---

🌐 FastAPI Deployment

▶️ Run the API Server

cd app
uvicorn app:app --reload

Open UI
http://127.0.0.1:8000/docs

Sample Request
{
  "features": [30 numerical features]
}


Sample Response
{
  "fraud_probability": 0.99,
  "prediction": "Fraud"
}


🏆 Key Highlights

✔️ Highly imbalanced classification handled correctly

✔️ Recall-optimized model tuning

✔️ Fully production-ready REST API

✔️ Business-impact-based evaluation

✔️ Industry-level project structure
=======
# credit-card-fraud-detection-ml
>>>>>>> 819d3375d770bc08d214beacd924b60d0ab33961