Variance Cut

When growing a regression tree (predicting a continuous target) rather than a classification tree, which criterion is typically used to choose the best split at each node?

The split that minimizes the number of distinct target values in each child node

The split that minimizes the weighted sum of variance (or SSE) of the target within each resulting child node

The split that maximizes Gini impurity reduction across child nodes

The split that maximizes the entropy of the target distribution in each child node