Concepts de base
Grafting consistent estimators onto Random Forests ensures consistency and improved performance.
Résumé
The article explores the consistency of Random Forests and introduces the concept of grafting consistent estimators onto a shallow CART. It discusses the shortcomings of traditional Random Forests and the benefits of the proposed approach. The paper includes theoretical results, empirical applications, experiments, and conclusions.
Introduction
- Little known about Random Forest theory
- Question of Random Forest algorithm consistency
Theoretical Results
- Consistency guarantee of the proposed approach
- Comparison with traditional Random Forests
Empirical Application
- Application to Boston Housing dataset
- Performance comparison with traditional Random Forests
Experiments
- L2 error analysis with increasing sample size
- Role of parameter αn in consistency
- Feature selection using CART step
Conclusion and future work
- Grafted Trees offer consistency and improved performance
- Potential for causal relationship extraction
Stats
Random Forests는 높은 차원 데이터에 적합
Citations
"Random Forests perform well in empirical settings."
"Grafted Trees outperform Centered Forests."