Why do Decision Trees tend to overfit small datasets?

Decision Trees overfit small datasets because they can create highly specific splits that perfectly memorize training samples.

With limited data, decision trees can continue splitting until each leaf contains very few samples or even a single sample. This leads to perfect training accuracy but poor generalization. The greedy algorithm prioritizes local impurity reduction, which becomes unreliable when data is sparse. Small datasets amplify variance, making trees highly unstable. Regularization techniques like pruning, min_samples_leaf, and limiting depth are essential in such cases.