Build a Bayesian Credit-Scoring Model for an Emerging-Markets Fintech
Overview
What this challenge is about.
You receive an anonymized snapshot of about 30,000 historical applications with features (income proxy, tenure on platform, prior loans, region) and the binary default outcome. Build a Bayesian logistic regression with weakly-informative priors using PyMC or NumPyro. Compare against the incumbent gradient-boosted model on: (a) ROC-AUC, (b) expected calibration error, (c) credible-interval width per applicant, and (d) demographic-parity gap across two region groups. Recommend whether the risk team should A/B test the Bayesian model in production for thin-file applicants only.
The Brief
What you'll do, and what you'll demonstrate.
Determine whether a Bayesian credit model improves calibration and fairness on thin-file applicants without sacrificing too much discrimination.
Earning criteria — what you'll demonstrate
- Apply Bayesian inference to a real risk-modeling problem
- Quantify and compare model calibration with appropriate metrics
- Reason about fairness/discrimination/calibration trade-offs in credit
- Translate a Bayesian result into a risk-team-actionable recommendation
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Data Scientist
Building a calibrated, fairness-aware credit model and recommending an A/B test scope is the canonical fintech data scientist's project.
This challenge sharpens
- bayesian-learning
- credit-scoring
- fairness-metrics
ML Researcher
Posterior-predictive checks and pre-registered fairness slices map to the research-methodology rigor expected in industry ML research.
This challenge sharpens
- bayesian-learning
- calibration
- model-evaluation