The decision tree regression predicts true value of an instance from the average of the training labels at the corresponding decision node. The decision threshold is determined from the minimum quadratic sum of the label distance. Let’s try a decision tree depth of 3 again.
val reg = new DecisionTreeRegressor(depth=3)
This is the application of a decision tree regression algorithm to the cubic dataset. The prediction is quite crude.
Increasing the tree depth increases the number of nodes by about factors of two. With this setting, the decision tree is able to fit the test data reasonably well.
val reg = new DecisionTreeRegressor(depth=6)
