- Concepts
- Keeping up with ML innovation
- Machine learning
- Supervised learning
- Predictive model
- Tree ensemble
- Random forest
Random forest
Model familyRandom forests combine trees trained with randomized data or feature choices to reduce correlated prediction errors.
In plain terms: A random forest trains many decision trees on random subsets of data and features and averages their votes, so their individual mistakes cancel out.
For example: A forest of 500 trees, each seeing a bootstrap sample and a random third of the features, often beats a single tree by several points of accuracy on tabular data.
Why it matters
- Randomness in training examples and split candidates creates diverse tree predictions.
- Averaging correlated errors reduces variance less than averaging otherwise comparable uncorrelated errors.
- The random-input forest is one important formulation, not the only possible randomized-tree ensemble.
Common trap: A forest is not a single-tree subtype.
Read
Papers
- Random Forests (2001) (opens in a new tab)Springer Nature
About this resource
Combines randomized tree predictors and analyzes generalization and out-of-bag evaluation; the ensemble extends single-tree ideas.
Where to read: Original random-forest construction and out-of-bag analysis
Checked 2026-10-08
Authority for this purpose: Existing scholarly source independently inspected against this concept; publication category is kept distinct from preprint and instruction.
Destination inspection: Bounded GET; public DNS/IP pinned; redirect-by-redirect validation; destination text inspection required; reviewed destination title, provenance and concept-specific subject
Useful destination; no explanatory claim is asserted from this link.
- Useful scoped source; attaching it does not establish historical priority or independent superiority.
Attached reference
Original reference type: scholarly publication
Original reference verification scope: Bibliographic eligibility only; see scoped relationship evidence.
Foundational reference in the ontology
Relationship: hasFoundationalSource · outgoing
The checked primary account supplies earlier or foundational work relevant to this concept under the stated scope.
Relationship scope: Formalizes a broad randomized-tree ensemble family and develops random-feature and bagging-based variants. The paper explicitly acknowledges Ho, Amit and Geman, and earlier ensemble approaches. Do not claim exclusive invention of all randomized tree ensembles. No exclusive-inventor or universal earliest-priority claim is made.
Relationship review status: reviewed source-role attribution
Where to read: Author-hosted January 2001 manuscript, section 1.1 and definition 1.1; sections 3–3.1.
Open the cited evidence (opens in a new tab)
Original reference type: scholarly publication
Original reference verification scope: Bibliographic eligibility only; see scoped relationship evidence.
A selected publication link does not by itself establish original invention or verify every claim.
Reviewed authoritative resource
Where to read: Original random-forest construction and out-of-bag analysis
A first-party destination supports its documented scope; it does not establish independent performance validation or historical priority.
Existing link correction: Scoped resource correction or accessible alternative after actual destination review; historical academic evidence is not edited.
- Breiman: Random Forests (2001), author-hosted manuscript (opens in a new tab)Leo Breiman; University of California, Berkeley
About this resource
Combines randomized decision-tree predictors and analyzes prediction error, feature selection and out-of-bag evaluation; the manuscript version is distinct from the journal record.
Where to read: Original January2001 random-forest author manuscript
Checked 2026-10-08
Authority for this purpose: Existing scholarly source independently inspected against this concept; publication category is kept distinct from preprint and instruction.
Destination inspection: Bounded GET; public DNS/IP pinned; redirect-by-redirect validation; destination text inspection required; reviewed destination title, provenance and concept-specific subject
Useful destination; no explanatory claim is asserted from this link.
- Useful scoped source; attaching it does not establish historical priority or independent superiority.
Attached reference
Original reference type: academic paper copy
Original reference verification scope: Bibliographic eligibility only; see scoped relationship evidence.
Reviewed authoritative resource
Where to read: Original January2001 random-forest author manuscript
A first-party destination supports its documented scope; it does not establish independent performance validation or historical priority.
Existing link correction: Scoped resource correction or accessible alternative after actual destination review; historical academic evidence is not edited.
Courses
- James, Witten, Hastie and Tibshirani: An Introduction to Statistical Learning, first edition, seventh printing (opens in a new tab)Gareth James, Daniela Witten, Trevor Hastie and Robert Tibshirani / official author-hosted edition
About this resource
Provides introductory supervised-learning, regression/classification and tree-method instruction; the older edition’s retrieval restrictions remain explicit.
Where to read: Statistical Learning: requested supervised-learning and tree chapters
Checked 2026-10-08
Authority for this purpose: Existing scholarly source independently inspected against this concept; publication category is kept distinct from preprint and instruction.
Destination inspection: Bounded GET; public DNS/IP pinned; redirect-by-redirect validation; destination text inspection required; reviewed destination title, provenance and concept-specific subject
Useful destination; no explanatory claim is asserted from this link.
- Useful scoped source; attaching it does not establish historical priority or independent superiority. Original book publisher was not established by this unavailable PDF retrieval. Destination inspection is unverified; this record is retained explicitly and does not count as an accessible verified source.
Attached reference
Original reference type: academic book, paper or university instructional text
Original reference verification scope: Bibliographic eligibility only; see scoped relationship evidence.
Reviewed authoritative resource
Where to read: Statistical Learning: requested supervised-learning and tree chapters
A first-party destination supports its documented scope; it does not establish independent performance validation or historical priority.
Existing link correction: Scoped resource correction or accessible alternative after actual destination review; historical academic evidence is not edited.
Related concepts
Broader
- Tree ensembleModel family
Narrower
- Quantile regression forestModel family
- Random feature subsampling at splitsMethod
Related
- Compare trees, bagging, forests and boostingSkill
- Decision treeModel family
- BaggingMethod
- scikit-learnTool
- ClassificationTask
- RegressionTask
- Out-of-bag evaluationMethod
Papers and people
Books to go deeper
Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow
Aurélien Géron · 3rd edition
Build and evaluate predictive models and practical preprocessing pipelines.
View on Amazon (affiliate link) ↗Probabilistic Machine Learning: An Introduction
Kevin P. Murphy · 2022 edition
Probability, inference and predictive modeling with uncertainty.
View on Amazon (affiliate link) ↗As an Amazon Associate I earn from qualifying purchases.
Loading discussion…
