# Overfitting **Domain:** Machine Learning / Statistics **Doc Type:** Canonical Concept Node **Maturity:** Foundational ## Definition **Overfitting** occurs when a model captures peculiarities of its training data so closely that performance fails to generalize to new observations. ## Evaluation Context [[wiki/Held-Out Validation|Held-out validation]] tests for this failure by separating data used for fitting from data used for evaluation. Regularization, capacity control and broader sampling address related aspects of the problem. ## Key Insight **Excellent performance on remembered examples is not the same as transferable capability.** ## See Also [[wiki/Training Data|Training Data]], [[wiki/AI Benchmarking|AI Benchmarking]], [[wiki/Machine Learning|Machine Learning]]