Principle:Scikit learn contrib Imbalanced learn Balanced Random Forest
| Knowledge Sources | |
|---|---|
| Domains | Machine_Learning, Ensemble_Learning, Imbalanced_Learning |
| Last Updated | 2026-02-09 03:00 GMT |
Overview
An ensemble technique that trains each decision tree on a balanced bootstrap sample by drawing equal numbers of majority and minority class instances.
Description
Balanced Random Forest modifies the standard Random Forest by altering the bootstrap sampling at each tree. Instead of drawing a standard bootstrap from the full dataset, it draws a bootstrap sample from the minority class and then samples with replacement the same number of instances from the majority class. Each tree is trained on this balanced subset. The ensemble aggregates predictions through voting.
This approach is more effective than applying resampling before training a standard Random Forest because each tree sees a different balanced view of the data, increasing diversity.
Usage
Use this principle when:
- A Random Forest is the desired model family
- Per-tree balancing is preferred over global resampling
- Ensemble diversity matters for generalization
- Computational cost of training is acceptable
Theoretical Basis
For each tree t in the ensemble:
- Draw a bootstrap sample of size n from the minority class
- Sample n instances with replacement from the majority class
- Train decision tree t on the combined balanced sample
The final prediction is the majority vote across all trees.