Build an MLP Baseline for Credit-Default Risk at a Fintech
Overview
What this challenge is about.
You receive 18 months of anonymized credit-decision data (around 600,000 applications, 80 features) with a 90-day default label. Train an MLP with regularization (dropout, weight decay), early stopping, and class-balanced sampling. Compare to a baseline XGBoost on AUC, calibration, KS statistic, and predict-then-cutoff approval-rate-at-default-rate. Address explainability with SHAP. Deliverable is the trained MLP, the comparison memo, and an explainability appendix.
The Brief
What you'll do, and what you'll demonstrate.
Match or beat XGBoost on AUC, calibration, and approval-rate-at-default-rate using a single regularized MLP.
Earning criteria — what you'll demonstrate
- Apply regularization (dropout, weight decay) on tabular MLPs
- Compare deep models against strong tree baselines fairly
- Evaluate calibration on a credit-risk model
- Communicate model behavior to a CRO audience
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.
Career paths this builds toward
Canonical rolesMachine Learning Engineer
Replacing tree baselines with calibrated MLPs and writing the CRO memo is exactly the kind of first project a junior MLE owns at a fintech.
This challenge sharpens
- mlp
- regularization
- calibration
Data Scientist
Model comparison with calibration and SHAP-based explanations is a canonical credit-risk data-scientist deliverable.
This challenge sharpens
- calibration
- shap
- tabular-deep-learning
Applied AI Scientist
Translating deep-learning parity into a CRO sign-off package mirrors the applied-AI-scientist's bridging role.
This challenge sharpens
- mlp
- calibration
- shap