When a statistical model fits its training data's noise, not just its true pattern.
Overfitting occurs when a model or prediction rule is tuned so closely to the quirks of one dataset that it captures random noise alongside genuine signal, inflating apparent accuracy. Such models tend to perform much worse on new patients or datasets than their original validation statistics suggest, a gap sometimes called 'optimism.' It is a particular risk when a study has many candidate predictors relative to its sample size.
This guide was auto-drafted and is pending editorial review.