AI/ML
Credit Card Fraud Detection MLOps
July 17, 2026
This project takes fraud detection past the notebook stage into something closer to production. It trains a RandomForestClassifier on the classic (and severely imbalanced 492 fraud cases out of 284,807 transactions) Kaggle credit card dataset, favoring precision/recall/PR-AUC over raw accuracy given the class imbalance.
Data and model artifacts are versioned with DVC instead of committed to git, experiment runs are logged to MLflow, and a FastAPI service exposes /predict for inference. GitHub Actions runs the test suite and a Docker build on every push using a synthetic placeholder model in CI so the pipeline doesn't depend on a private DVC remote.
Why: Everyone trains a model and calls it done. I wanted to go further build the full pipeline around it: an API to serve it, and DVC to actually version the data and model artifacts.
Stack: scikit-learn, MLflow, DVC, FastAPI, Docker, GitHub Actions