Overfitting

Overfitting is fitting or selecting a model around development-data features that do not generalize to the intended population. It can arise from repeated model selection even when the final model has few parameters.

Overfitting occurs when fitting or selecting a model exploits idiosyncrasies of the development data that do not generalize to the intended population. Repeatedly choosing among many variants can overfit even if the final model looks simple. Parameter count is one influence, not a complete diagnosis.

Evaluation must preserve the separation between development choices and the evidence used to assess them. A repeatedly consulted test set becomes part of development. Poor future performance can also reflect distribution shift or a changed target rather than overfitting alone, so diagnose the source before applying simplicity or regularization as a universal cure.

When to use it

When analyzing data; when building predictive models; when evaluating statistical claims; when quantifying uncertainty.

How it can help

Define the generalization target, separate development and evaluation, account for the full selection procedure, and inspect leakage or distribution change alongside model complexity.