Fraud Detection
End-to-end MLOps pipeline for credit card fraud detection
About
Complete machine learning operations pipeline for detecting fraudulent credit card transactions. Implements the full ML lifecycle from data ingestion and feature engineering through model training, evaluation, and deployment. Uses PCA for dimensionality reduction on high-dimensional transaction features, with Random Forest and Logistic Regression as competing models tracked via MLflow experiment management.
The production-ready REST API serves predictions through FastAPI, all containerized with Docker for reproducible deployments. The pipeline automates model comparison and promotes the best-performing model to the serving endpoint, ensuring consistent and reliable fraud detection in production.
Key Features
- Random Forest & Logistic Regression with PCA dimensionality reduction
- MLflow experiment tracking and model versioning
- FastAPI REST API for real-time predictions
- Docker containerized for reproducible deployments
- Automated model evaluation and comparison
- Feature engineering pipeline for transaction data
Architecture
Three-tier architecture: data processing layer (scikit-learn pipelines with PCA), model training layer (MLflow-tracked experiments comparing RF and LR), and serving layer (FastAPI endpoints). Docker Compose orchestrates the full stack including the MLflow tracking server and API service.