Decision Tree (CART)

A decision tree built with CART (Classification and Regression Trees) recursively splits the data on predictor values to form a tree of if-then rules that predicts a categorical or continuous outcome.

CART(분류·회귀나무)로 만드는 의사결정나무는 예측변수 값을 기준으로 자료를 재귀적으로 분할하여 범주형 또는 연속형 결과를 예측하는 조건 규칙의 나무를 만든다.

How do I build an easy-to-explain set of rules that predicts an outcome?

결과를 예측하는 설명하기 쉬운 규칙을 어떻게 만드나요?

A CART decision tree splits the data step by step on the predictor values that best separate the outcome, producing readable if-then rules. It fits classification or regression when interpretability matters, with cost-complexity pruning used to limit overfitting.

When to use Decision Tree (CART)

• A categorical or continuous outcome with mixed predictor types
• Rules that non-specialists can read are needed
• Exploring interactions and thresholds in predictors

When to use something else

• Maximum predictive accuracy; use random forest or gradient boosting
• Smooth linear effects and coefficient inference; use linear or logistic regression

Assumptions and requirements

• Labelled training data
• Pruning or depth limits to control overfitting
• Validation on held-out data
• Trees are unstable under small data changes

What the analysis reports

• Tree diagram with split rules
• Variable importance
• Cross-validated error and complexity parameter
• Accuracy, confusion matrix or RMSE on test data

How to report it

The pruned CART tree had [value] terminal nodes and achieved a test-set accuracy of [value]%.

Related methods

• Random Forest
• Conditional Inference Tree
• Gradient Boosting Machine

Run Decision Tree (CART) on Quorant