Why are Decision Trees sensitive to small changes in training data?

Decision Trees are sensitive because each split depends on the dataset distribution, and small changes can alter the entire tree structure.

Decision Trees are hierarchical and greedy. A small change in the dataset can change the best split at the root node. Since all downstream splits depend on earlier decisions, this change propagates through the entire tree, producing a completely different structure. This property leads to high variance and instability, especially in deep trees. Ensemble methods like bagging reduce this instability by averaging multiple trees.