Decision trees partition the feature space into a set of hypercubes, and then fit a simple model in each hypercube. The simple model can be a prediction model, which ignores all predictors and predicts the majority (most frequent) class (or the mean of a dependent variable for regression), also known as 0-R or constant classifier.

Decision tree induction forms a tree-like graph structure as shown in the figure below, where:

A test is a rule for partitioning the feature space. A test depends on feature values. Each outcome of a test represents an appropriate hypercube associated with both the test and one of the descending branches.

If a test is a Boolean expression (for example, or , where

f

is a feature and

c

is a constant fitted during decision tree induction), the inducted decision tree is a binary tree, so its non-leaf nodes have exactly two branches, ‘true’ and ‘false’, each corresponding to the result of the Boolean expression.

Prediction is performed by starting at the root node of the tree, testing features by the test specified in this node, then moving down the tree branch corresponding to the outcome of the test for the given sample. This process is then repeated for the subtree rooted at the node, discovered at the selected branch. The final result is the prediction of the simple model at the leaf node.

Decision trees are often used in ensemble algorithms, such as boosting, bagging, or decision forest.