Conceptdata leakage
Dataset Leakage
A common problem in which training data inadvertently contains information from the test or validation set, causing inflated model evaluation results and poor generalization ability.
Timeline (last 90 days)
Oct 22
Measures to prevent dataset leakage include strictly splitting data, checking for duplicate samples, and avoiding global normalization before splitting
Unverified50%