What is typically chosen for the minimum number of observations in each terminal node of a regression tree?

In the context of regression trees, a small minimum number of observations in each terminal node is typically chosen to allow for more granular splits in the data. This practice helps capture the underlying patterns within the data more effectively, particularly in heterogeneous datasets where relationships may change across different segments.

By allowing a small number of observations, the tree-building process can identify more detailed variations and potentially enhance predictive accuracy. However, it's important to balance this with the risk of overfitting, where the model become too tailored to the training data, capturing noise rather than general trends.

Choosing a small minimum number fosters greater sensitivity to data variations, enabling the model to recognize unique influences specific to smaller groups within the dataset, which can lead to better decision-making when applied to unseen data.