Stack Five Models for a Kaggle-Style Forecasting Bake-Off
Overview
What this challenge is about.
You receive a pseudonymized dataset of 24 months of daily shipment volumes across about 200 origin-destination lanes plus weather and holiday features. Train 5 base models, use 5-fold time-series cross-validation, then train a stacking meta-learner (ridge or lightgbm) on out-of-fold predictions. Report mean absolute percentage error (MAPE) per lane segment (high-volume vs. long-tail) and per horizon (day-1 vs. day-7). Write a 4-page post-mortem that explains the trade-offs and names the single model the team should ship if they could only deploy one.
The Brief
What you'll do, and what you'll demonstrate.
Beat a moving-average baseline on next-day shipment forecasting across heterogeneous lanes and horizons, using a 5-model stacked ensemble.
Earning criteria — what you'll demonstrate
- Apply ensemble methods (stacking) to a real forecasting problem
- Engineer time-series cross-validation correctly (no future leakage)
- Reason about the heterogeneity of forecast error across segments
- Communicate forecasting trade-offs in a post-mortem format
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
A clean ensemble forecasting project with honest per-segment reporting is the portfolio piece that gets a junior data scientist past the screen at most industry teams.
This challenge sharpens
- ensemble-methods
- time-series-forecasting
- cross-validation
Machine Learning Engineer
Productionizing a stacked ensemble with reproducible code and clear evaluation is the day-one MLE shape.
This challenge sharpens
- ensemble-methods
- lightgbm
- scikit-learn