Using a New Set of Cases
Examples:
- Cost-complexity pruning: model the predicted error rate as the weighted sum of its complexity and its error on the training cases, with the separate cases used primarily to determine an appropriate weighting
- Reduced-error pruning: assess the error rates of the tree and its components directly on the set of separate cases
Drawback:
- some of available data must be reserved for the separate set, so the original tree must be constructed from a smaller training set