CMStatistics 2022: Start Registration
View Submission - CMStatistics
B0480
Title: Construction of random forest based on parametric bootstrapping Authors:  Asanao Shimokawa - Tokyo University of Science (Japan) [presenting]
Etsuo Miyaoka - Tokyo University of Science (Japan)
Abstract: Ensemble learning improves prediction accuracy by combining multiple learners. Random forest is one of the typical methods of ensemble learning, and it combines the tree-structured models as the individual learners. In many studies, individual learners are trained based on samples obtained by sampling with replacement of supervised data. Although this is a non-parametric bootstrap-based learning method, if there is additional information about the population distribution, it seems more plausible to apply the parametric bootstrap method. Therefore, we focus on random forest based on parametric bootstrap method and compare it with the conventional ensemble learning method through theoretical and simulation studies. In addition to this, we show the results of the forest obtained from actual data.