The paper delves into the theory behind Random Forests, addressing their inconsistencies and limitations. It introduces the concept of grafting consistent estimators onto CART trees, demonstrating improved performance and adaptability in empirical studies. The study highlights the importance of consistency for inference, especially in applications involving causal relationships. Various variants of the algorithm are discussed, focusing on Centered Trees and Kernel Regression. The empirical application on the Boston Housing dataset showcases the superior performance of Grafted Trees over traditional methods like Breiman's Random Forest. Experiments on simulated data further validate the effectiveness of Grafted Trees in different scenarios. Feature selection and consistency considerations are also explored, emphasizing the potential benefits of using this approach for predictive modeling.
Til et andet sprog
fra kildeindhold
arxiv.org
Vigtigste indsigter udtrukket fra
by Nicholas Wal... kl. arxiv.org 03-12-2024
https://arxiv.org/pdf/2403.06015.pdfDybere Forespørgsler