A decision tree is based on a binary tree datastructure and applies a one dimensional cut-off based on one feature at each node. The cut-off threshold and the feature it is applied to are chosen to maximize a purity metric for the resulting split in the training data. The depth of the tree sets the number of binary decisions to be taken.
val clf = new DecisionTreeClassifier()
This is the application of a decision tree algorithm to the circular dataset. In the default configuration the number of decisions is not sufficient to appropriately detect the circle:
Increasing the tree depth roughly doubles the number of nodes. With this setting, the decision tree is able to classify the test data reasonably well.
val clf = new DecisionTreeClassifier(depth=4)
