Machine Learning / MLOps
Fraud Detection MLOps.
Version data and models, track experiments and use a model registry with MLflow and DVC.
Problem
Make a fraud detection experiment traceable and reproducible.
Context
Educational ML lifecycle project using synthetic transactions.
Objective
Track experiments, version data and manage model versions.
Dataset / Data
Synthetic educational transaction dataset. No production banking data.
Architecture
DVC / Git versioned data → Random Forest → MLflow Tracking → Model Registry.
Methodology
MLflow experiment “tp1-decouverte”. Used a registered model and production alias concepts.
My contribution
Project reported in the candidate’s background. Individual scope and team responsibilities remain to be confirmed.
Challenges
Connect data versions, experiments and registered models.
Limitations
Synthetic data. No production deployment or real-world detection improvement is claimed.
What I learned
ML lifecycle traceability, registries and data / model versioning.
Project results
Synthetic transactions
Educational dataset; 568 fraudulent transactions. No validated predictive metric supplied.