一种顺序响应的随机森林:变量预测和选择

发布时间：2019-05-09 22:37

【摘要】：利用随机森林法(RF)进行分类和回归,在过去被广泛地研究,然而在基于顺序响应的情况下并没有标准的方法.在随机森林(RF)的基础上通过广泛的研究,对条件推理树进行探索,以探讨结合顺序信息是否改善预测性能或提高变量选择的效果.本文提出的两种置换变量的重要性预测方法 RPS-VIM和MAE-VIM经过实验验证是优化的方法,它替代目前存在的内置变量重要性测量方法 ER-VIM和MSE-VIM.基于真实数据的结果表明在某些设置中,有序回归树中使用RPS-VIM和MAEVIM顺序响应的组合,预测的排名可以得到改善,并且预测精度优于原始的基于分类树的随机森林.
[Abstract]:Random forest method (RF) is used for classification and regression, which has been widely studied in the past, but there is no standard method in the case of sequential response. On the basis of random forest (RF), the conditional reasoning tree is explored through extensive research, in order to explore whether the combination of sequential information can improve the prediction performance or improve the effect of variable selection. The importance prediction methods RPS-VIM and MAE-VIM proposed in this paper are verified by experiments as optimization methods, which replace the existing built-in variable importance measurement methods ER-VIM and MSE-VIM.. The results based on real data show that in some settings, the rank of prediction can be improved by using the combination of RPS-VIM and MAEVIM sequential response in the ordered regression tree, and the prediction accuracy is better than that of the original random forest based on classification tree.
【作者单位】：兰州交通大学电子与信息工程学院;
【基金】：国家自然科学基金项目(61163010)资助甘肃自然科学基金项目(1308RJZA111)资助兰州市科技计划项目(2015-2-99)资助
【分类号】：O212

【相似文献】