Pre-pruning decision trees
When building a decision tree to avoid overfitting you may want to stop the branching process even if it could increase the accuracy on the training data. One method to do this is to put in requirements on the final form of the decision tree before you start training. The requirements might be:
- Max depth of the decision tree in any one branch.
- Minimum size of a leaf node for training data.
- Minimum size of a parent node for training data.