随机森林算法中数据切分方法研究  

Research on Data Segmentation Method in Random Forest Algorithm

作　　者：路佳佳 LU Jia-jia(Department of Computer Information Engineering,Shanxi Institute of Business and Technology,Taiyuan 030006,China)

机构地区：[1]山西工商学院计算机信息工程学院,山西太原030006

出　　处：《唐山师范学院学报》2021年第3期73-75,共3页Journal of Tangshan Normal University

摘　　要：应用随机切分方法将模拟数据集切分成互不相交的两部分,然后从这两部分中进行P次抽样得到2P份数据集,最后从2P份数据集中随机选择一份建立随机森林。实验结果表明,经切分处理的随机森林在回归任务中偏差更小,在分类任务中准确率更高。当ntree在100左右时,误差就趋于稳定。以iris数据集为处理对象,对上述方法进行了验证,结果表明,经切分处理的数据集得到的方差更加稳定。The simulation data set is divided into two disjoint parts by using the random segmentation method,and then 2p data sets are obtained by p times sampling from the two parts.Finally,a random forest is established by randomly selecting one of the 2p data sets.The experimental results show that the segmented random forest has smaller deviation in regression task and higher accuracy in classification task.When the number of tree is about 100,the error tends to be stable.Taking iris data set as the processing object,the above method is verified,and the results show that the variance obtained from the segmented data set is more stable.

关 键 词：随机森林 随机抽样 误差分析 切分方法 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]