Fraud Detection Model 🔍

A production-ready machine learning system for detecting fraudulent credit card transactions using XGBoost, MLflow, and MinIO.

📊 Model Performance

Metric	Score	Status
Accuracy	99.87%	🏆 Excellent
AUC	99.79%	🏆 Excellent
Precision	93.28%	✅ Very Good
Recall	83.21%	⚠️ Good (Improving)
F1 Score	87.96%	✅ Very Good

🚀 Quick Start

Prerequisites

• Docker & Docker Compose
• Python 3.9+
• Git

1. Clone & Setup

git clone https://github.com/your-username/transfraud-model.git
cd transfraud-model

2. Start Services

# Start MLflow, MinIO, and monitoring services
docker-compose up -d

3. Install Dependencies

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r requirements.txt

4. Prepare Data

# Create data directory and add your dataset
mkdir -p data
# Place your credit-card-transaction-fraud-dataset.csv in data/ folder

5. Train Model

python src/train.py

🏗️ Project Architecture

transfraud-model/
├── src/                    # Source code
│   ├── train.py           # Main training script
│   ├── data_utils.py      # Data loading & validation
│   ├── features.py        # Feature engineering
│   ├── eval_utils.py      # Model evaluation
│   └── __init__.py
├── data/                   # Dataset directory
├── artifacts/             # Generated artifacts
│   ├── models/            # Trained models
│   ├── plots/             # Evaluation plots
│   ├── reports/           # Performance reports
│   └── analysis/          # Feature analysis
├── logs/                  # Training logs
├── docker-compose.yml     # Service orchestration
├── requirements.txt       # Python dependencies
└── README.md             # This file

🛠️ Tech Stack

Core ML

• XGBoost - Gradient boosting for fraud detection
• Scikit-learn - Data preprocessing & evaluation
• Pandas & NumPy - Data manipulation

MLOps & Infrastructure

• MLflow - Experiment tracking & model registry
• MinIO - S3-compatible artifact storage
• Docker - Containerization
• Boto3 - AWS SDK for MinIO integration

Monitoring & Visualization

• Matplotlib/Seaborn - Performance visualization
• Structured Logging - Comprehensive training logs

📈 Model Features

Feature Engineering

• Time-based Features: Hour, day, month extraction
• Location Features: Distance calculations
• Categorical Encoding: Merchant, category, location encoding
• Transaction Patterns: Amount analysis, frequency features

Model Optimization

• Class Imbalance Handling: Scale positive weight adjustment
• Threshold Optimization: Dynamic classification thresholds
• Multiple Model Strategies: Balanced, recall-optimized, high-recall variants

🔧 Configuration

Environment Variables

MLFLOW_TRACKING_URI=http://localhost:5001
MLFLOW_S3_ENDPOINT_URL=http://localhost:9000
AWS_ACCESS_KEY_ID=minio
AWS_SECRET_ACCESS_KEY=minio123

Docker Services

• MLflow: http://localhost:5001
• MinIO: http://localhost:9000
• MinIO Console: http://localhost:9001

📊 Monitoring & Evaluation

MLflow Tracking

Access experiment results at: http://localhost:5001

• Parameter tracking
• Metric comparison
• Artifact storage
• Model versioning

MinIO Artifacts

Access stored models at: http://localhost:9001

• Organized model versions
• Evaluation plots
• Performance reports
• Feature analysis

Generated Artifacts

• ROC Curves & Confusion Matrices
• Feature Importance Analysis
• Classification Reports
• Model Performance Metrics

🎯 Model Training

Basic Training

python src/train.py

Advanced Options

# With debug logging
python src/train.py --log-level DEBUG

# Single model strategy (faster)
python src/train.py --single-model

Training Process

1. Data Validation - Schema and quality checks
2. Feature Engineering - Time, location, categorical features
3. Model Selection - Multiple XGBoost variants compared
4. Threshold Optimization - Recall-precision balance
5. Artifact Generation - Plots, reports, analysis
6. Model Registration - MLflow model registry
7. Storage - MinIO artifact organization

🗂️ Data Requirements

Expected Dataset Format

amt,is_fraud,merchant,category,city,state,lat,long,merch_lat,merch_long,trans_date_trans_time
29.99,0,store_A,gas,New York,NY,40.7128,-74.0060,40.7130,-74.0062,2023-01-01 08:30:00

Required Columns

• amt: Transaction amount (numeric)
• is_fraud: Fraud indicator (0/1, binary)
• merchant, category: Transaction details
• Location fields: lat, long, merch_lat, merch_long
• Timestamp: trans_date_trans_time

🔍 Model Interpretation

Key Features

The model prioritizes these patterns for fraud detection:

1. Transaction Amount Patterns
2. Geographic Anomalies
3. Time-based Suspicious Activity
4. Merchant Category Risk Levels
5. User Behavioral Patterns

Business Impact

• 83% Fraud Detection Rate (Recall)
• 93% Alert Accuracy (Precision)
• <7% False Positive Rate
• 99.8% Overall Accuracy

🚨 Alerting & Monitoring

Performance Thresholds

Minimum Standards:
  - Recall: >80%
  - Precision: >85%
  - F1 Score: >85%
  - AUC: >95%

Monitoring Dashboard

Access real-time metrics at:

• MLflow: http://localhost:5001
• MinIO: http://localhost:9001

🔄 CI/CD Pipeline

Automated Workflow

1. Data Validation → 2. Model Training → 3. Evaluation → 4. Registry → 5. Deployment

Quality Gates

• Data quality checks
• Model performance validation
• Artifact integrity verification
• Storage confirmation

🛡️ Production Readiness

✅ Completed

• Model training pipeline
• Experiment tracking
• Artifact storage
• Model versioning
• Performance monitoring
• Docker containerization

🔄 In Progress

• Real-time inference API
• Automated retraining
• Advanced feature engineering
• A/B testing framework

🤝 Contributing

Development Setup

1. Fork the repository
2. Create feature branch: git checkout -b feature/amazing-feature
3. Commit changes: git commit -m 'Add amazing feature'
4. Push to branch: git push origin feature/amazing-feature
5. Open Pull Request

Code Standards

• Follow PEP 8 guidelines
• Include type hints
• Add comprehensive logging
• Update documentation

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Dataset: Kaggle Fraud Detection
• MLflow for experiment tracking
• XGBoost for robust gradient boosting
• MinIO for S3-compatible storage

---

Ready for Production 🚀

For questions or support, please open an issue or reach out to me at nestorabiawuh@gmail.com.