Random Forests
Random Forests
复制标题
DOI：
10.1007/978-0-387-77501-2_5
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Richard A. Berk
中科院分区：
文献类型：
--
作者：
Richard A. Berk
本章继续建立在统计学习程序的集合的想法。介绍了随机森林，这是一种非常有用的方法，扩展和改进了装袋。和以前一样，有一个分类或回归树的集合，并对树进行投票以进行正则化。当在每个树的每个潜在分区处，选择预测的随机子集进行评估时，引入了额外的随机性。这有各种各样的好处，其中一些可能是非常微妙的。还讨论了随机森林的补充算法，允许窥视黑盒。
This chapter continues to build on the idea of ensembles of statistical learning procedures. Random forests is introduced, which is an extremely useful approach that extends and improves on bagging. As before, there is an ensemble of classification or regression trees and votes over trees to regularize. Additional randomness is introduced when at each potential partitioning for each tree, a random subset of prediction is selected for evaluation. This has a variety of benefits, some of which can be quite subtle. Also discussed are supplementary algorithms to random forests that allow a peek into the black box.