Recursion: Build the Full Tree

Learn how decision trees repeat the split search on each impure child until the tree is complete.

Finding the root split is only the first step. A decision tree grows by recursion—repeating the same split search on every impure child until the tree is finished.

Big idea

After the root split:

  • A pure child (all one class) becomes a leaf with a prediction.
  • An impure child is treated like a new mini-dataset—the algorithm runs the same steps again: generate thresholds, score weighted Gini, pick the best split.

This repeats down the tree until every branch ends in a leaf.

Formula

At each impure node, the tree applies the same rule as the root:

Best split=argminfeature,t  Weighted Gini\text{Best split} = \arg\min_{\text{feature},\, t} \;\text{Weighted Gini}

Recursion stops at a node when:

  • the node is pure (Gini =0= 0), or
  • a stopping condition applies (covered in a later lesson)

Worked example

On our 8-sample practice dataset, the root split is Feature 0 2.5\leq 2.5:

BranchSamplesResult
Right (x>2.5x > 2.5)4All class 1 → leaf, predict 1
Left (x2.5x \leq 2.5)4Still mixed → recurse

On the left branch, the tree searches again among the remaining features and thresholds. It may pick Feature 0 1.5\leq 1.5 next, creating one pure leaf and one branch that still needs splitting.

Eventually every path ends in a leaf—some after one split, others after several.

Common mistake

Thinking the root split alone is the full tree. The root is just the first decision; recursion on impure children builds the rest.

Key takeaway

Decision trees grow by recursion: find the best split at each impure node, create leaves where groups are pure, and repeat until stopping rules say stop.

Try tracing recursive splits in the practice notebook below.

Practice

Try It Yourself

Open the practice lab to complete the starter code in the notebook.

Knowledge Check

Quick Quiz

After the root split, what happens to an impure child node?

Summary

Key Takeaways

  • After the root split, impure child nodes are split again using the same process.
  • Pure child nodes become leaves with a class prediction.
  • Recursion continues until all leaves are pure or stopping rules apply.