Choose a Parameter to Reduce Underfitting in Gradient-Boosted Trees

# Choose a Parameter to Reduce Underfitting in Gradient-Boosted Trees A gradient-boosted tree model is underfitting. Which parameter could you decrease to allow a more flexible fit: tree depth, number of leaves, the proportion of observations used to fit a tree, or the minimum number of observations per leaf? Explain why your choice can help and how you would verify the diagnosis. ### What a Strong Answer Covers - The role of a minimum leaf-size constraint in limiting tree flexibility. - Why reducing depth or leaf count usually increases capacity constraints. - The uncertainty of changing row-sampling fraction and the need for training/validation evidence. ```hint Consider a blocked split A useful split may be forbidden if one child contains fewer observations than the minimum leaf size. ``` ### Follow-up Questions - What risk rises when the minimum leaf size becomes very small? - How would you distinguish underfitting from poor features or a target-definition error?