Generalization error

From Machinelearning
Jump to: navigation, search


The generalization error of a model (with particular chosen values of the parameters of the model) is the expected value of the cost function of the model on a test set that has been completely withheld from the process used to train the model.

Note that generalization error is closely related to test set error and cross-validation set error. Explicitly the generalization error gives the expected value of both the test set error and the cross-validation set error, though the actual values of the errors on these sets would depend on the specific choice of the test set and cross-validation set.