💳 Credit Card Fraud Detection

A full end-to-end machine learning pipeline for detecting fraudulent credit card transactions using anomaly detection (Isolation Forest, LOF) and supervised learning (XGBoost), served via an interactive Streamlit dashboard.

---

📁 Project Structure

fraud-detection/
│
├── data/                        # Raw and processed datasets
│   └── .gitkeep
│
├── models/                      # Saved model artifacts
│   └── .gitkeep
│
├── notebooks/
│   └── fraud_detection.ipynb    # Full exploratory + training notebook
│
├── src/
│   ├── __init__.py
│   ├── data_loader.py           # Dataset download & loading
│   ├── preprocessor.py          # Scaling, SMOTE balancing
│   ├── anomaly_detection.py     # Isolation Forest & LOF
│   ├── classifier.py            # XGBoost training & evaluation
│   ├── visualizer.py            # ROC curve, confusion matrix plots
│   └── predictor.py             # Inference on new transactions
│
├── tests/
│   └── test_pipeline.py         # Unit tests
│
├── app.py                       # Streamlit dashboard entry point
├── train.py                     # Full training pipeline script
├── requirements.txt
└── README.md

---

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Download the Dataset

Get the dataset from Kaggle: Credit Card Fraud Detection

Place creditcard.csv in the data/ directory.

Or use the Kaggle API:

pip install kaggle
kaggle datasets download -d mlg-ulb/creditcardfraud -p data/ --unzip

3. Train the Models

python train.py

6. Launch the Dashboard

streamlit run app.py

---

📊 Dataset

• Source: Kaggle - ULB Credit Card Fraud
• Size: 284,807 transactions
• Fraud Rate: ~0.17% (highly imbalanced)
• Features: 30 (V1–V28 PCA features + Time + Amount)

---

🧠 Models Used

Model	Type	Purpose
Isolation Forest	Unsupervised	Anomaly detection
Local Outlier Factor	Unsupervised	Anomaly detection
XGBoost Classifier	Supervised	Final classification

---

⚖️ Handling Class Imbalance

• SMOTE (Synthetic Minority Oversampling Technique) is applied on the training set
• StandardScaler is used to normalize Amount and Time
• Stratified train/test split preserves fraud ratio

---

📈 Evaluation Metrics

• Confusion Matrix
• ROC-AUC Curve
• Precision, Recall, F1-Score
• Average Precision Score

---

🖥️ Dashboard Features

• Upload or enter transaction data manually
• Real-time fraud probability prediction
• Anomaly score visualization
• Model performance comparison charts
• Interactive ROC curve

---

🧪 Running Tests

python -m pytest tests/

---

📦 Requirements

See requirements.txt for full list. Key packages:

• pandas, numpy
• scikit-learn
• xgboost
• imbalanced-learn (SMOTE)
• streamlit
• matplotlib, seaborn, plotly
• joblib

---