Data & Datasets · Level 2 of 5
Data Leakage
Information entering training or development that would be unavailable at real prediction time.
Leakage can inflate evaluation results without improving deployable performance.
Example
A predictor accidentally uses a field recorded only after the outcome.
Listen to the definition and example
Audio transcript
Data Leakage. Information entering training or development that would be unavailable at real prediction time. Leakage can inflate evaluation results without improving deployable performance. For example: A predictor accidentally uses a field recorded only after the outcome.
Explore this concept
Why it matters
This helps you prevent misleading experiments caused by data preparation and split mistakes.
Start with
Related concepts
Quick recall question
Try answering before looking back at the definition.