EasyDeepLearn

Gini impurity, entropy, and classification error — which do trees actually use?

medium

Answer

  • For classification, trees choose the split that most reduces impurity.
  • Gini = 1 - sum pc2p_{c}^{2} and entropy = -sum pcp_{c} log pcp_{c} are almost interchangeable in practice, both being smooth and differentiable proxies for classification error.
  • Gini is slightly faster (no log) and is CART's default.
  • Classification error is not smooth enough — it's rarely used to grow trees, only to prune them.
  • For regression, the criterion is variance reduction (MSE).
Check yourself — multiple choice
  • Classification error is the standard splitting criterion
  • Gini and entropy are the practical choices; classification error is used only for pruning
  • Regression trees use Gini
  • Entropy is not differentiable

Gini/entropy for growth (smooth); classification error for evaluation/pruning.

#decision-trees#trees

Practise Supervised Learning

215 interview questions in this topic.

Related questions