Fig. 1.
Examples of model fitting to data. The data are synthetic, for the purposes of illustration only. (A) An underfit representation of the data. Although the linear discriminator captures most of the green circles, numerous red crosses are misclassified. The linear model is too simple. (B) The discriminator is overfit to the data. Although there are no classification errors for the example data, the model will not generalize well when applied to new data that arrives. (C) A parabola discriminates the data appropriately with only a few errors. This is the best parsimonious classification.