Detect Fraudulent Refund Requests for a Mid-Market Marketplace
Overview
What this challenge is about.
Train two calibrated classifiers to detect fraudulent refund requests for a marketplace and recommend an operating threshold. You get a verifiable certificate.
The scenario
The marketplace (Series C, around 250 staff, gross merchandise value about USD 380M annually) currently loses an estimated USD 1.4M per year to this fraud pattern and operates with a 4-person trust & safety review team.
The Brief
What you'll do, and what you'll demonstrate.
Build a calibrated fraud-detection model whose operating point matches the trust & safety team's manual-review capacity.
Earning criteria — what you'll demonstrate
- Handle a heavily imbalanced classification problem with appropriate techniques
- Use probability calibration (Platt scaling, isotonic regression) and reliability diagrams
- Tie operating points to real operational constraints
- Communicate trade-offs to a non-technical operations team
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Aligned coursework coming soon.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
- Classification
Apply classification to solve real industry problems and demonstrate production-level capability.
- Model Calibration
Apply model calibration to solve real industry problems and demonstrate production-level capability.
- Imbalanced Classification
Apply imbalanced classification to solve real industry problems and demonstrate production-level capability.
- Feature Engineering
Apply feature engineering to solve real industry problems and demonstrate production-level capability.
- Model Evaluation
Apply model evaluation to solve real industry problems and demonstrate production-level capability.
- Python
Write clean, efficient Python for data processing, automation, and backend services.
Careers
Career paths this challenge builds toward
Completing this challenge demonstrates skills that transfer directly to these roles:
Data Scientist
Calibrated classifiers tied to operational capacity are exactly the kind of work that junior data scientists own at marketplaces, fintechs, and trust & safety teams.
This challenge sharpens
- classification
- model-calibration
- model-evaluation
Applied AI Scientist
Choosing operating points based on real ops capacity instead of pure metrics is what separates applied work from research and is a daily applied-AI-scientist task.
This challenge sharpens
- model-calibration
- imbalanced-classification
- model-evaluation
AI Engineer
Packaging a calibrated model with a defensible threshold recommendation is the kind of glue work AI engineers do when handing models to operations teams.
This challenge sharpens
- python
- feature-engineering
- model-calibration