Transaction Fraud Detector

An end-to-end MLOps pipeline for detecting fraudulent credit card transactions.

An end-to-end MLOps project covering the full lifecycle from raw data to a served model with monitoring, automated retraining, and deployment gating. Built in 10 stages, it includes data validation, XGBoost training with MLflow experiment tracking, automated model promotion gating, REST API serving with FastAPI, shadow mode, canary deployment simulation, A/B testing, drift detection, and automated retraining pipelines.

Tech Stack

PythonXGBoostMLflowFastAPIEvidentlyDockerScikit-learnGitHub Actions

Features

XGBoost model training with MLflow experiment tracking
Automated model promotion gating with champion/challenger comparison
REST API serving with FastAPI and Pydantic validation
Shadow mode, canary deployment, and A/B testing with statistical significance
Data drift detection with Evidently and rolling performance monitoring
Automated retraining pipeline triggered by drift or degradation
CI/CD with GitHub Actions
Dockerized deployment

Challenges

Handling extreme class imbalance in fraud detection
Building a robust promotion gate with threshold checks
Implementing canary and shadow deployment strategies

Feedback

For feedback or suggestions, contact me at: dev.jhawar.cs@gmail.com

Link

Github