What is true about decision tree pruning?

Decision tree pruning is an essential technique in tree-based modeling to reduce overfitting and improve the model's generalization to unseen data.

The first statement about recursive binary splitting leading to overfitting is accurate. When a decision tree is grown without constraints, it can continue to split the data in a way that captures noise rather than the underlying trend, resulting in a model that performs very well on training data but poorly on test data.

The second statement indicating that a tree with more splits has higher variance is also correct. More splits typically result in a model that becomes overly sensitive to variations in the training set, leading to different predictions for slightly different data points. This increased sensitivity contributes to a higher variance.

The third statement regarding cost complexity pruning states that when the complexity parameter α is set to zero, it allows the decision tree to grow to its maximum size, resulting in a very large tree. This is true, as a zero-penalty for complexity means that no pruning occurs, and thus, the tree retains all its splits, potentially leading to overfitting.

Given that both the first and third statements are valid, the choice that includes them is indeed the correct selection. Pruning is a crucial aspect of tree-based methods like decision trees