How to Build a Decision Tree on Mac & Windows

A decision tree is the most readable predictive model there is — a flowchart of yes/no questions. Here's how to grow one on Mac & Windows and keep it honest.

A decision tree predicts an outcome by asking a sequence of simple questions about the predictors — “is price above X? is the area below Y?” — splitting the data at each step. The result is a model you can read like a flowchart, which is why it's so popular when interpretability matters.

Grow it

Open your data and choose Decision Tree.
Pick the outcome (a category for classification, a number for regression) and the predictors.
Read the tree — each split is the question that best separates the data at that point.

A decision tree in Stratum — Each split is a yes/no question; follow the branches to a prediction.

Keep it from overfitting

Left unchecked, a tree will keep splitting until it memorizes the training data — and then predicts new data poorly. The fix is pruning: Stratum uses cross-validation to cut the tree back to the size that generalizes best, so you get a model that's accurate and readable.

Which factors matter? A variable-importance panel ranks the predictors by how much they improve the splits — a quick read on what's actually driving the outcome.

Predict with it

Add a predicted-column back to your data in one click. And when you need more accuracy than a single tree, ensembles — random forests and boosted trees — combine many trees; see Lesson 24. Full walk-through in Lesson 23.

Download Stratum

Frequently asked questions

What is a decision tree used for?

Predicting an outcome — a category (classification) or a number (regression) — through a readable sequence of yes/no splits. It's prized when you need to explain why a prediction was made.

How do I stop a decision tree from overfitting?

Prune it. Stratum uses cross-validation to cut the tree back to the size that predicts new data best, rather than letting it memorize the training set.

What's the difference between a decision tree and a random forest?

A decision tree is one readable model; a random forest averages many trees for higher accuracy at the cost of interpretability. Start with a tree to understand the data, move to a forest for prediction.

Can a decision tree handle both categories and numbers?

Yes — classification trees predict a category, regression trees predict a number, and both can use numeric and categorical predictors.