Fit Regimes

Overfitting is associated with high variance: the model picks up noise in the training data and therefore fails to generalise well to new, unseen data. Underfitting is associated with high bias: the model is too simple to capture the true relationship underlying the data.