Calibrate a Demand Forecast with Bayesian Confidence Intervals
Overview
What this challenge is about.
You receive 24 months of weekly demand for 600 SKUs plus the existing XGBoost point predictions. Fit a Bayesian conformal-prediction layer (or, alternatively, a Gaussian-Process residual model) that converts point predictions into 80 percent intervals per SKU. Validate calibration on the most recent 12 weeks using empirical coverage and the Continuous Ranked Probability Score (CRPS). Success is empirical coverage between 76 and 84 percent on the holdout, and a 1-page operations brief that converts the intervals into per-SKU safety-stock recommendations.
The Brief
What you'll do, and what you'll demonstrate.
Add a calibrated probabilistic layer over an existing point-forecast model so operations can size safety stock per SKU with documented confidence.
Earning criteria — what you'll demonstrate
- Apply Bayesian and conformal methods for prediction-interval estimation
- Validate interval calibration with empirical coverage and proper scoring rules
- Translate probabilistic outputs into operational inventory decisions
- Communicate uncertainty to non-statistician operators
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
Probabilistic forecasting plus operational translation is the bread and butter of data-scientist roles at consumer brands and retailers.
This challenge sharpens
- bayesian-inference
- uncertainty-quantification
- operations-translation
Machine Learning Engineer
Wrapping an existing model with a calibrated probabilistic layer and validating it on a rolling holdout mirrors common MLE add-on work.
This challenge sharpens
- python
- model-calibration
- conformal-prediction
Applied AI Scientist
Choosing between Bayesian and conformal approaches based on the operational need is the kind of judgement applied AI scientists exercise weekly.
This challenge sharpens
- bayesian-inference
- conformal-prediction
- model-calibration