Dear friends, for the regression trees on numerical features, how are we deciding when to split a feature?
Say X = [X_1 | X_2 | … | X_n]
and the DT decides to pick X_1 based on variance reduction, how is the rule X_1 <= a decided? In other. words, how to choose a?
Many thanks
Dear Alex,
Both the node or attribute on which to split and the point (a) on where to split is decided based on brute force. For each split in each attribute, the variance is calculated to get the reduction. Thereafter, the highest one is chosen among all attributes and splits.