Data Leakage
Domain: Data & Datasets
🎯 Motivation
When test data accidentally leaks into training data. Causes inflated scores that do not hold in real use.
📋 Concrete Example
Leakage: the same article appears in both train and test sets. The model memorizes it. Test score: 95%. Real-world score: 70%. The leakage caused false confidence.
🔗 Analogy
A student accidentally seeing the exam before the test - they score 95% but do not really know the material. Data leakage gives false high scores.