Credit Card Fraud Detection
End-to-end MLOps pipeline for financial security.
The Problem
Financial institutions lose billions to fraud annually. The challenge is balancing high recall (catching all fraud) with high precision (not blocking legitimate users).
Key Decisions
- ZenML for Orchestration: Used to manage the end-to-end workflow, ensuring reproducibility from raw data to batch inference.
- RobustScaler & OneHot Encoding: Crucial preprocessing decisions to handle the heavy outliers and categorical variance typical in financial transaction data.
- Streamlit Dashboard: Built to provide human analysts with fraud probability scoring and risk classification, making the model’s output actionable.
Results
- Outcome: A deployable fraud detection system with real-time risk classification.
- Actions: Automated recommended business actions (Allow/Review/Block) based on score thresholds.
What I Learned
Handling imbalanced datasets requires more than just a good model; it requires a robust preprocessing pipeline and a deep understanding of the business cost of false positives vs false negatives.
[Built as an MLOps MVP. Source code is public.]