A Machine Learning Project Checklist

Before you trust a result

Is the question precise, with a metric tied to real costs? Is there a baseline? Was the test set split off first and used only once? Is preprocessing inside a pipeline? Have you checked for leakage and suspicious features? Are results reported with cross-validation spread? Are classification thresholds chosen on validation data? Have you looked at errors and per-group performance? Is the model saved with its data version and metrics? Is there a monitoring and retraining plan?

The checklist

Use it in reviews.

[ ] question + metric linked to real costs
[ ] baseline (dummy model / current rule)
[ ] test set split first, used once
[ ] preprocessing in a pipeline
[ ] leakage check: every feature available at prediction time?
[ ] CV mean and spread reported
[ ] threshold chosen on validation data
[ ] error analysis + per-group results
[ ] model saved with data version and metrics
[ ] monitoring + retraining plan

Explain it to a non-expert

If you cannot explain what the model predicts, how well, and when it fails in plain words, it is not ready.

Quick check: Which item belongs on an ML project checklist?

• Skip the baseline
• Tune on the test set
• Split off the test set first and use it only once
• Fit scaling on all data before splitting