05
Domain 05
Trees, Forests & Boosting
Splits and depth, bagging versus boosting, early stopping, feature importance, and where gradient boosting still wins.
6 questions
- Q25What is happening, and which tree should ship?99.6% train, 74% test. Which tree ships?
- Q26Why does the random feature subset at each split improve the forest?Why a forest deliberately hides features from its own trees.
- Q27What should the analyst do?Validation loss bottoms out at tree 500. Training loss keeps falling. Now what?
- Q28Which importance should be trusted, and why?The #1 feature is a customer ID hash. Which importance lied?
- Q29Based on the 2022–2026 benchmark evidence, which outcome is most likely?GBDT vs neural net vs tabular foundation model on 40k rows — the 2026 answer.
- Q30Which statement correctly contrasts the two?Parallel or sequential, variance or bias: bagging vs boosting in one line.