How do I recognize overfitting?

Training error is markedly lower than test (or cross-validated) error. The bigger the gap, the worse the generalization.

Is a high R² a sign of a good model?

Not necessarily — with enough predictors you reach a high R² even on random data. Adjusted R² and error on new data decide.

Overfitting

Definition

Overfitting happens when a model learns not only the real relationships but also the random noise of the training data — it looks great on them and fails on new data. It is typical of complex models (deep trees, large networks, regression with many predictors) and small datasets. It shows up as a gap between training and test error.

Data mining & machine learning

Defences: more data, a simpler model, regularization (penalizing complexity), tree pruning, early stopping of network training, variable selection and above all honest validation on data the model never saw during training. The opposite is underfitting — a model too simple to capture the relationship.

Overfitting threatens classical statistics too: a regression with 20 predictors on 30 observations has a high R² but zero explanatory value. As a rule of thumb you need at least 10–20 observations per estimated parameter.

In Statistica

Against overfitting Statistica offers test sets and v-fold cross-validation in all data-mining modules, tree pruning by validated error, automatic stopping of boosting and networks by test error, and stepwise selection or information criteria in regression; the Workspace environment shows training and test error side by side.

Available from edition: Statistica Modeler

Related terms

Knowledgebase guides

Statistica Workspace – graphical user interface

FAQ

How do I recognize overfitting?: Training error is markedly lower than test (or cross-validated) error. The bigger the gap, the worse the generalization.
Is a high R² a sign of a good model?: Not necessarily — with enough predictors you reach a high R² even on random data. Adjusted R² and error on new data decide.

Try it on your own data

Statistica free for 30 days

Full version, no credit card. Or get a pricing estimate in a minute.

Try free for 30 days Pricing estimate

Data mining & machine learning

Decision trees Random forests and boosting Neural networks Cross-validation ROC curve and AUC Model scoring and deployment Text mining All terms →

Updated: September 2026.