Decision trees select the feature and its split threshold of a node based on the sum of errors of their child leaves

At any given node, a decision tree will choose two things:

• the feature to split on
• the threshold to split on It calculates and sums errors on both sides of the split and chooses the cheapest option.