# Decision Tree Regressor: domain of the y variable

just wondering about a thing. suppose you fit a Decision Tree Regressor and your training y variable has got a domain that spans from 0 to 1 included, is it possible that the predicted y variable of the fitted model on the test set to exceed these boundaries?

## 1 Answer

Decision Tree Regressor won't predict values outside the range of values they are given in the train set. If your extreme values are -4 and +10 the predicted values will be between -4 and +10. The reason for that is that a decision tree splits the training data in groups. The prediction associated with a node is then the mean value of the training data of that node (or if you choose something else than MSE, whatever value that minimize the cost function for this group).

An example of how a regression tree works : https://www.saedsayad.com/decision_tree_reg.htm The output for Windy and Sunny is 26.5 as the training data has 2 data points that fit these criteria (23 and 30)

• Excellent thank you! Sep 19 '20 at 12:23