Understanding Random Forests
summary
Starts from CART and implements decision trees with entropy and information gain, then actually ensembles them with bagging and per-tree feature subsampling to make Random Forests; uses scikit-learn on real datasets (Etsy ~60k sellers, UCI bank) and reports that modest numbers of estimators already smooth predictions, reduce overfitting, and give usable out-of-bag error estimates; highlights key knobs (n_estimators, max_features, min_samples_leaf) and where forests beat or lose to simple linear models.
AI-written summary · about
chapters (8)
- 0:00Random Forests intro
- 7:32Decision trees and overfitting
- 15:40Etsy data example
- 23:25Single tree limitations
- 26:00Why forests improve trees
- 31:23Forest parameters
- 35:39Forest advantages
- 40:20When to use forests
related talks
57mFrom Bonsais to Random Forests
27mUsing Random Forests in Python
40mCART: Not only Classification and Regression Trees
36mAdventures in scikit-learn's Random Forest
53mA Taste of Random Decision Forests on Apache Spark
06mPërdorimi i Decision Trees dhe të dhënat për të marrë vendime
54mDebiasing Random Forests For Treatment Effect Estimation
43mFeature Importance and Ensemble Methods : a new perspective