What is the role of probability estimation in Decision Trees?

Decision Trees estimate probabilities based on class frequencies in leaf nodes.

In classification trees, probability estimates are derived from the proportion of training samples belonging to each class in a leaf node. For example, if a leaf has 80 positive and 20 negative samples, the probability for the positive class is 0.8. However, these estimates can be poorly calibrated, especially in small leaves. Techniques like Laplace smoothing or ensemble averaging improve probability calibration.