ML EngineeringOpen
Predict churn without cheating
Four columns leak the answer. Find them, drop them, and still beat the honest baseline.
- Time needed
- 2 weeks
- Data
- 310k customers
The brief
Most churn submissions score beautifully and mean nothing, because a payment-status column already encodes the answer. This challenge is scored on a held-out future window, so leakage does not help you. You will need to frame the split by time, justify your features, and explain your model's predictions.
What you submit
- A leak-free pipeline with a time-based split
- PR-AUC on the held-out window plus your threshold choice and its business rationale
- SHAP-based explanation of the top drivers
How to participate
- Register free: create your account and tell us a little about yourself.
- Download the dataset and the leakage warning in the brief.
- Train, evaluate on the time-based split, and explain your model.
- Submit your repo and scores; mentors re-run your pipeline.
How it’s judged
- No leakage
- PR-AUC on held-out window
- Quality of explanation
- Threshold reasoning
What you walk away with
Feedback for every entry · Top 3 win a full flagship seat
Resources
Everything you need, in one zipThe dataset, the full brief and the guides for this challenge.Register free to unlock